Imaginal Discs: The Genetic and Cellular Logic of Pattern Formation (Developmental and Cell Biology Series)

IMAGINAL DISCS With the elucidation of the complete ﬂy genome, traditional ﬂy genetics is in more demand than ever. Gene...

Author: Lewis I. Held | Jr.

53 downloads 1519 Views 5MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

IMAGINAL DISCS With the elucidation of the complete ﬂy genome, traditional ﬂy genetics is in more demand than ever. Genetics will allow us to explain the role of each of the 14,000 genes, many of which are involved in the development of imaginal discs. These hollow sacs of cells make adult structures during metamorphosis, and their study is crucial to comprehending how a larva becomes a fully functioning ﬂy. This book examines the genetic circuitry of the well-known “fruit ﬂy,” tackling questions of cell assemblage and pattern formation, of the hows and the whys behind the development of the ﬂy. The book ﬁrst establishes that ﬂy development relies primarily on intercellular signaling, and then discusses how this signaling occurs. After an initial examination of the proximity versus pedigree imperatives, the book delves into bristle pattern formation and disc development, with entire chapters devoted to the leg, wing, and eye. Extensive appendices include a glossary of protein domains, catalogs of well-studied genes, and an outline of signaling pathways. More than 30 wiring diagrams, among 67 detailed schematics, clarify the text. The text goes beyond the Internet databases insofar as it puts these myriad facts into both a conceptual framework and a historical context. Overall, the aim is to provide a comprehensive reference guide for students and researchers exploring this fascinating, but often bewildering, ﬁeld. Lewis I. Held, Jr., is Associate Professor in the Department of Biological Sciences at Texas Tech University.

Developmental and Cell Biology Series series editors Jonathan B. L. Bard, Department of Anatomy, Edinburgh University Peter W. Barlow, Long Ashton Research Station, University of Bristol David L. Kirk, Department of Biology, Washington University The aim of the series is to present relatively short critical accounts of areas of developmental and cell biology where sufﬁcient information has accumulated to allow a considered distillation of the subject. The ﬁne structure of cells, embryology, morphology, physiology, genetics, biochemistry and biophysics are subjects within the scope of the series. The books are intended to interest and instruct advanced undergraduates and graduate students and to make an important contribution to teaching cell and developmental biology. At the same time, they should be of value to biologists who, while not working directly in the area of a particular volume’s subject matter, wish to keep abreast of developments relevant to their particular interests. books in the series 18. C. J. Epstein The Consequences of Chromosome Imbalance: Principles, Mechanisms and Models 19. L. Sax´en Organogenesis of the Kidney 20. V. Raghavan Developmental Biology of the Fern Gametophytes 21. R. Maksymowych Analysis of Growth and Development in Xanthium 22. B. John Meiosis 23. J. Bard Morphogenesis: The Cellular and Molecular Processes of Developmental Anatomy 24. R. Wall This Side Up: Spatial Determination in the Early Development of Animals 25. T. Sachs Pattern Formation in Plant Tissues 26. J. M. W. Slack From Egg to Embryo: Regional Speciﬁcation in Early Development 27. A. I. Farbman Cell Biology of Olfaction 28. L. G. Harrison Kinetic Theory of Living Pattern 29. N. Satoh Developmental Biology of Ascidians 30. R. Holliday Understanding Ageing 31. P. Tsonis Limb Regeneration 32. R. Rappaport Cytokinesis in Animal Cells 33. D. L. Kirk Volvox: Molecular Genetic Origins of Multicellularity and Cellular Differentiation 34. R. L. Lyndon The Shoot Apical Meristem: Its Growth and Development 35. D. Moore Fungal Morphogenesis 36. N. Le Douarin & C. Kalcheim The Neural Crest, Second Edition 37. P. R. Gordon-Weeks Neuronal Growth Cones 38. R. Kessin Dictyostelium 39. L. I. Held, Jr. Imaginal Discs: The Genetic and Cellular Logic of Pattern Formation

0 521 25464 7 0 521 30152 1 0 521 33022 X 0 521 35327 0 0 521 35053 0 0 521 43612 5 0 521 36115 X 0 521 24865 5 0 521 40943 8 0 521 36438 8 0 521 30691 4 0 521 35221 5 0 521 47802 2 0 521 44149 8 0 521 40173 9 0 521 45207 4 0 521 40457 6 0 521 55295 8 0 521 62010 4 0 521 44491 8 0 521 58364 0 0 521 58445 0

Thomas Hunt Morgan (3rd from right) and his associates at Columbia University. This luncheon was held in the “Chart Room” on 2 January 1919, to celebrate the return of Alfred Henry Sturtevant (foreground with beer and cigar) from his brief stint as a soldier in World War I [72, 651, 1556, 2283]. Calvin Bridges (center) is feigning a chat with a museum mannequin (Homo erectus) dressed in Sturt’s uniform. Clockwise from this anthropoid “guest” are Hermann J. Muller, T. H. Morgan (“the Boss”), Frank E. Lutz, Otto L. Mohr, Alfred F. Huettner, A. H. Sturtevant, Franz Schrader, Ernest G. Anderson, Alexander Weinstein, S. C. Dellinger, and Calvin B. Bridges. Curt Stern (not shown) did not join the team until 1924 [3071]. This merry band of pioneers launched a great quest for the secrets of genetics, and they had a knack for solving mysteries that rivaled Sherlock Holmes [72, 650, 651, 2951, 4182, 4184]. Nevertheless, as the informality of this party indicates, these legendary heroes did not take themselves too seriously [72, 3903]. Indeed, their lightheartedness has suffused this ﬁeld ever since [4696] and is reﬂected in the whimsical names of many ﬂy genes [2561]. Most of the mutations they studied affect the adult’s anatomy by altering the development of the larva’s imaginal discs. Those discs are the subject of this book, one of whose aims is to celebrate the triumph of the quest. This picture is from Sturt’s photo album. It was provided courtesy of the Archives, California Institute of Technology.

IMAGINAL DISCS The Genetic and Cellular Logic of Pattern Formation

LEWIS I. HELD, JR. Texas Tech University

CAMBRIDGE UNIVERSITY PRESS Cambridge, New York, Melbourne, Madrid, Cape Town, Singapore, São Paulo Cambridge University Press The Edinburgh Building, Cambridge CB2 2RU, UK Published in the United States of America by Cambridge University Press, New York www.cambridge.org Information on this title: www.cambridge.org/9780521584456 © Lewis I. Held, Jr. 2002 This book is in copyright. Subject to statutory exception and to the provisions of relevant collective licensing agreements, no reproduction of any part may take place without the written permission of Cambridge University Press. First published 2002 This digitally printed first paperback version 2005 A catalogue record for this publication is available from the British Library Library of Congress Cataloguing in Publication data Held, Lewis I., 1951– Imaginal discs : the genetic and cellular logic of pattern formation / Lewis I. Held, Jr. p. cm. Includes bibliographical references and index. ISBN 0-521-58445-0 1. Drosophila melanogaster – Genetics. 2. Drosophila melanogaster – Embryology. 3. Drosophila melanogaster – Morphogenesis – Molecular aspects. 4. Cellular signal transduction. 5. Cell interaction. I. Title. QH470.D7 H45 2002 576.5 – dc21 2001043553 ISBN-13 978-0-521-58445-6 hardback ISBN-10 0-521-58445-0 hardback ISBN-13 978-0-521-01835-7 paperback ISBN-10 0-521-01835-8 paperback

Contents

Preface

CHAPTER ONE. CELL LINEAGE VS. INTERCELLULAR SIGNALING

Discs are not clones No part of a disc is a clone, except claws and tiny sense organs Cells belong to lineage “compartments”

CHAPTER TWO. THE BRISTLE

Numb segregates asymmetrically and dictates bristle cell fates Delta needs to activate Notch, but not as a signal per se Amnesic cells can use sequential gating to simulate a binary code Notch must go to the nucleus to function E(spl)-C genes are Su(H) targets but play no role in the SOP lineage The transcription factor Tramtrack implements some cell identities Hairless titrates Su(H) Several other genes help determine the 5 cell fates Pox neuro and Cut specify bristle type Bract cells are induced by bristle cells Macrochaetes and microchaetes differ in size but not in kind

CHAPTER THREE. BRISTLE PATTERNS

Surprisingly, different macrochaete sites use different signals Prepatterns may contain hidden “singularities” How Achaete and Scute control bristles was debated for decades In 1989, Achaete and Scute were found to mark “proneural clusters” In 1995, the old AS-C paradigm toppled and a new one emerged Proneural “spots” shrink to SOP “dots” The SOP uses a feedback loop to raise its Ac and Sc levels Two other bHLH genes (asense and daughterless) assist SOPs

page xi

1 1 4 4

5 5 9 10 12 15 18 20 23 27 28 29

31 33 36 36 43 44 45 48 48

vii

viii

CONTENTS

“Lateral” or “mutual” inhibition ensures one SOP per PNC Notch-pathway and proneural genes are functionally coupled Doses of Notch-pathway genes can bias the SOP decision Extra SOPs could be inhibited by contact or diffusion (or both) Scabrous may be the diffusible SOP inhibitor Inhibitory ﬁelds dictate the spacing intervals of microchaetes Microchaetes come from proneural stripes, not spots Hairy paints “antineural” stripes on the legs Leg bristles use extra ﬁne-tuning tricks Chemosensory leg bristles are patterned like notal macrochaetes Extramacrochaetae superimposes an uneven antineural “mask” Dose dependency implies that HLH proteins “compute” bristles Robustness of patterning may be due to a tolerant time window Atonal and Amos are proneural agents for other types of sensilla Other (upstream) pathways govern bristle patterning

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

Segmentation genes set the stage for disc initiation Prepatterns and gradients clashed in trying to explain homeosis Homeotic genes implement regional identities Wing and haltere discs “grow out” from 2nd- and 3rd-leg discs Thoracic discs arise at Wingless/Engrailed boundaries Cell lineage within compartments is indeterminate The Polar Coordinate Model linked regeneration to development But regeneration has peculiarities that set it apart

CHAPTER FIVE. THE LEG DISC

The Molecular Epoch of disc research was launched in 1991 Bateson’s Rule (1894) governs symmetry planes in branched legs Meinhardt’s Boundary Model deftly explained Bateson’s Rule The Boundary and PC Models jousted in a “Paradigm War” Hh, Dpp, and Wg are the chief intercellular signaling molecules P-type cells use Hh to “talk” to A-type cells nearby Hh elicits expression of Dpp and Wg along the A/P boundary Dpp dorsalizes and Wg ventralizes, or do they? Dpp and Wg are mutually antagonistic Dpp and Wg jointly initiate distal outgrowth But Dpp seems more crucial than Wg as a growth factor The A/P boundary can migrate when its “jailors” are “asleep” Regeneration is due to a Hh spot in the peripodial membrane The Polar Coordinate Model died in 1999 How Hh, Dpp, and Wg move is not known, nor is their range Whether Dpp and Wg travel along curved paths is not known Hairy links global to local patterning Questions remain about the Hh-Dpp-Wg circuitry Distal-less is necessary and sufﬁcient for distalization

49 50 52 52 53 55 57 61 62 67 68 71 73 74 75

76 76 80 84 85 87 91 92 96

97 97 99 100 103 105 107 109 111 114 115 118 119 122 123 124 125 128 128 129

CONTENTS

Proximal and distal cells have different afﬁnities Dachshund is induced at the Homothorax/Distal-less interface Homothorax and Extradenticle govern the proximal disc region Fasciclin II is induced at the BarH1/Aristaless interface BarH1 and Bric a` brac affect P-D identity, joints, and folds Leg segmentation requires Notch signaling

CHAPTER SIX. THE WING DISC

The A-P axis is governed by Hh and Dpp but not by Wg Dpp turns on omb and spalt at different thresholds Dpp regulates omb and spalt similarly despite clues to the contrary Dpp does not regulate tkv in 3rd instar despite clues to the contrary A vs. P identities might explain how a straight A/P line emerges But the A/P line appears to straighten via a signaling mechanism Intercalation is due to a tendency of Dpp gradients to rise The variable height of Dpp gradients makes them appear seamless A Wg gradient speciﬁes cell fates along the wing’s D-V axis Perpendicular (Dpp × Wg) gradients suggest Cartesian coordinates But cells do not seem to record positional values per se Wg’s repression of Dfz2 is inconsequential Apterous’s role along the D-V axis resembles Engrailed’s A-P role Chip cooperates with Apterous, and “Dorsal wing” acts downstream Serrate and Delta prod Notch to evoke Wg at the D/V line Fringe prevents Notch from responding to Serrate The core D-V circuit plugs into a complex network The wing-notum duality is established by Wg and Vein But Vestigial and Scalloped dictate “wingness” per se Straightening of the D/V border requires Notch signaling and Ap Straightening of veins may rely on similar tricks Two cell types predominate in the wing blade: vein and intervein Veins come from proveins that look like proneural ﬁelds But the resemblance is only superﬁcial All veins use the EGFR pathway But interveins also use the EGFR pathway (at a later time) Veins 3 and 4 are positioned by the Hh pathway Veins 2 and 5 are positioned by the Dpp pathway The Dpp pathway later implements the vein state A cousin of Dpp (Gbb) fosters the A and P cross-veins Vein 1 uses a combination of Dpp and Wg signals Macrochaetes are sited by various “prepattern” inputs How bristle axons get wired into the CNS is not known

CHAPTER SEVEN. THE EYE DISC

Compound eyes have ∼750 facets, with 8 photoreceptors per facet Unlike the bristle, the ommatidium is not a clone The eye has D and V compartments (despite doubts to the contrary)

ix

132 132 133 134 135 135

137 137 140 143 143 148 149 153 155 156 157 158 158 158 160 161 164 165 167 171 173 173 174 175 175 177 184 185 186 188 188 189 190 191

197 197 201 202

x

CONTENTS

The Iroquois Complex controls D-V polarity via Fringe and Notch A morphogenetic wave creates the ommatidial lattice D-V polarity depends on a rivalry between R3 and R4 precursors R1–R8 cells arise sequentially, implying a cascade of inductions But the ﬁnal cell (R7) is induced by the ﬁrst one (R8) Various restraints prevent more than one cell from becoming R7 The information content of the inductive signals may be only 1 bit No transcription factor “code” has yet been found for R cells The lattice is created by inhibitory ﬁelds around R8 precursors The lattice is tightened when excess cells die Eye bristles arise independently of ommatidia The MF operates like a moving A/P boundary Dpp and Wg control the rate of MF progress The MF originates via different circuitry

CHAPTER EIGHT. HOMEOSIS

BX-C and ANT-C specify gross metameric identities along the body Ubx enables T3 discs to develop differently from T2 discs But Ubx does so by directly managing target genes in multiple echelons Pc-G and Trx-G “memory” proteins keep homeotic genes on or off Homothorax, Distal-less, and Spineless specify leg vs. antennal fates If a “master gene” exists for the eye, then it is also a micromanager The manifold “enhanceosome” is a wondrous Gordian Knot The deepest enigma is how evolution rewired the circuit elements

203 208 209 211 213 213 216 218 224 227 228 229 234 234

237 237 243 244 247 249 252 254 254

EPILOGUE

256

APPENDIX ONE. Glossary of Protein Domains

257

APPENDIX TWO. Inventory of Models, Mysteries, Devices, and Epiphanies

266

APPENDIX THREE. Genes That Can Alter Cell Fates Within the (5-Cell)

Mechanosensory Bristle Organ

271

APPENDIX FOUR. Genes That Can Transform One Type of Bristle Into

Another or Into a Different Type of Sense Organ

276

APPENDIX FIVE. Genes That Can Alter Bristle Number by Directly Affecting

SOP Equivalence Groups or Inhibitory Fields

278

APPENDIX SIX. Signal Transduction Pathways: Hedgehog,

Decapentaplegic, and Wingless

285

APPENDIX SEVEN. Commentaries on the Pithier Figures

297

References

307

Index

441

Preface

How embryos “self-assemble” has fascinated thinkers for millennia [2918, 3064, 3190]. Among the ancient Greeks, Aristotle (384–322 bce) made copious observations and coined the term “morphogenesis,” which is still in use today [2989, 4305]. For the past century, the science of “developmental mechanics” has hammered at this problem relentlessly, but it is only in the last decade that the core mysteries have ﬁnally cracked [1487]. The deepest secrets have come from a fairylike ﬂy named Drosophila melanogaster, probably the same species of “gnat” that Aristotle himself noticed hovering over vinegar slime [217, 3361, 4184]. Unfortunately, these insights can only be fully appreciated in the arcane language of ﬂy genetics. Hence this book full of runes and rules. This book concerns cuticular patterns, the cellular machinery that makes them, and the genetic circuitry that runs the machinery. Although it is mainly a survey, it is also a narrative that traces the roots of our knowledge. The story that it tells – albeit in condensed form – rivals the Iliad in scope (legions of researchers devoting decades to attacking thousands of genes) and the Odyssey in wonderment (monstrous mutants posing riddles that challenge even the most clever explorerheroes). Indeed, truth is often stranger than a fairy tale in the realm of the ﬂy. Believe it or not, there are even remote islands where giant drosophilids with dappled wings and feathery legs have been spied dancing and ﬁghting in the misty forests [668, 669]. Ever since 1910 when T. H. Morgan’s ﬁrst “ﬂy paper” was published [2948], the ﬁeld of ﬂy genetics has brimmed with intriguing curiosities [820, 2951, 3673] and equally colorful human personalities [120, 327, 2283, 4183]. Added to these delights is a menagerie of recently discovered molecules

– e.g., the midget “Bearded” (81 a.a.) [2499] and the giant “Dumpy” (3680 a.a.) [4668]. Now that the ﬂy genome project is ending [14], the world is peering into this circus. What newcomers may not realize is that this ﬁeld offers many diversions beyond its databases. Like other holometabolous insects, ﬂies live two lives – ﬁrst as a grub, then as a ﬂying adult [82]. During metamorphosis, 19 “imaginal discs” erupt from inside the maggot and are quilted together to form most of the adult skin. The gold-colored cuticle secreted by that skin is exquisitely ornate. The head is embossed with hundreds of domes that focus light onto bundles of photoreceptors, the thorax is sculpted into dozens of jointed parts that form a contraption for walking and ﬂying, and the abdominal wall (built from non-disc tissue [2648]) is pleated into an expansible chamber for digestion and reproduction. Nearly everywhere, the body surface sprouts bristles whose patterns can be as orderly as soldiers on parade. Why do only some cells make bristles? That is a problem of differentiation. Why do bristles arise only at certain sites? That is a problem of pattern formation, and these questions can be asked for structures in general. Beneath both problems is a coding enigma: how does the ﬂy’s 1-dimensional genome encode the 2dimensional cuticular landscape? Once, it seemed that each body part might be governed by its own set of genes [4509, 4512], but this notion proved wrong [1094, 1114, 2410, 4643]. In fact, most patterns are built by the same ensembles of genes. These modules arose eons ago in the mythical common ancestor of insects and vertebrates [1439, 3840]. Since then, evolution has customized the circuitry by making new intra- and inter-modular links [968, 1440].

xi

xii

What is the nature of the circuitry, and how does it program cells to “compute” patterns? That is the subject of this book. Topics are arranged roughly in order of increasing complexity. Chapter 1 establishes one simple fact: in contrast to nematodes, ﬂies rely primarily on intercellular signaling (vs. cell lineage) to assign cell fates. The rest of the book traces how signaling occurs. Chapter 2 delves into the 5-cell cluster that constructs a mechanosensory bristle. The bristle is an exception to the signaling rule: its cell fates are dictated almost entirely by lineage. Chapter 3 uses bristle patterns to show how cells communicate in populations larger than a bristle but smaller than a disc, and Chapter 4 sets the stage for a discussion of larger-scale patterning by reviewing how discs arise and grow. Chapters 5 to 7 explore how leg, wing, and eye discs use similar toolkits of genes in idiosyncratic ways. The other two major discs – haltere and genital – are excluded because their strategies so closely resemble wing [16, 51, 3875, 4683, 4684] and leg discs [679, 735, 1163, 2343, 2942, 3732], respectively. (Fly genitalia are evolutionarily modiﬁed appendages [1137, 1179, 1562].) Chapter 8 contemplates the phenomenon of homeosis in the context of evolution. Overall, the book’s quest is to understand cellular “epistemology” (what do cells know?) and “psychology” (how do they think?). Its approach involves de- and reconstruction: to cut through the jargon, tease out the facts, and then try to make sense of the models by piecing the clues back together using a priori reasoning. The bad news is that there are so many pieces in the puzzle that persistence will be needed. The good news is that their interactions are so limited that no fancy math is required to learn the rules of the game [3588, 3841]. A recurrent theme in the saga is how cellular riddles were solved by molecular genetics. The abiding moral is that there is much more experimental work to be done if we are to comprehend how the ﬂy’s ∼14,000 genes [14, 1559, 3618, 3674] – or a large portion thereof [280, 615, 963, 4273] – are orchestrated during patterning [698, 2162, 2237, 2845, 4084]. In short, the ﬂy still holds many secrets, and genomics will need genetics to ferret them out [465]. Thus, the book is a sampler of case studies and gedanken exercises, not an encyclopedia. That function is served by the Internet databases, and readers should consult two main websites: FlyBase ( ﬂybase.bio.indiana.edu) [124, 279] and The Interactive Fly (sdb.bio.purdue.edu) [484]. Fly lore is best savored by browsing the classics: the 1993 Cold Spring Harbor 2volume compendium on development [238], its gargan-

PREFACE

tuan 12-volume predecessor The Genetics and Biology of Drosophila [122], Mike Ashburner’s huge “handbook” [118], Lindsley and Zimm’s dictionary of ﬂy genes [2561], Bridges and Brehme’s Barnumesque catalog of freakish mutants [470], and the Morgan team’s magnum opus of 1925 [2951]. However, the fun of ﬂy research is best portrayed in the charming Fly by Martin Brookes (2001, Harper-Collins, N.Y.). Despite this disclaimer about breadth, a few topics are covered in depth in the appendices. Appendix 1 is a glossary of protein domains. Appendix 2 lists most of the ideas that have guided research in this ﬁeld. Appendices 3 to 5 catalog the well-studied genes that affect bristles, sensilla, or bristle patterns, and Appendix 6 outlines three of the key signaling pathways in disc development (Hedgehog, Wingless, and Decapentaplegic). The other two pathways are discussed in Chapters 2 (Notch) and 6 (EGFR). Appendix 7 contains additional comments about the ﬁgures. Historically, disc research has been reviewed intermittently. Disc histology was codiﬁed by Dietrich Bodenstein in 1950 [377]. Disc development and genetics were surveyed by Gehring and N¨othiger (1973) [1421], Postlethwait and Schneiderman (1973) [3448], Bryant (1978) [526], Shearn (1978) [3881], Poodry (1980) [3422], and Oberlander (1985) [3165]. The ﬁrst blush of moleculargenetic data was evaluated by Stephen Cohen in 1993 [834], and the fundamentals of signaling were summarized by Seth Blair in 1999 [358]. Two books that nicely bracket the last 30 years of investigation are The Biology of Imaginal Discs (1972, H. Ursprung and R. N¨othiger, eds.) [4426] and Developmental Genetics of Drosophila (1998, A. Ghysen, ed.) [1452]. Conventional nomenclature is used. Locations of genes are stated in terms of the salivary gland chromosome map [2561]: the 3-part code (e.g., “92E12–14”) denotes the chromosome section (1–20 span the X, 21–60 the 2nd, 61–100 the 3rd, and 101–102 the tiny 4th chromosome), the lettered subdivision (A–F), and the band or range of bands. Genes are italicized, but gene complexes (e.g., Bar-C) are not. All proteins are in plain type. Mutations are superscripted (e.g., numbLOF ), whereas wild-type alleles are not (numb) or are labeled with “+” (numb+ ). Null alleles are designated by a “null” or “−” superscript. Most gene names record the dominant (capital) or recessive (lowercase) nature of early mutations (e.g., Notch vs. numb). Capital “D” (Drosophila) is used for paralogs within the species (e.g., Dfz2 [310] in the frizzled series), whereas lowercase “d” refers to

PREFACE

orthologs of vertebrate genes (e.g., dTcf [692, 1517]). Proteins are always capitalized (e.g., Numb). Given these rules, the normal symbols for Hairless (H ) and hairy (h) are distinct for the genes but not for the proteins (“H” in both cases), so “H” will be used only for Hairless, while “Hairy” will be written out. Likewise, Beadex will be written out to avoid confusion with the protein encoded by bithorax (both would be “Bx”). Small capitals are employed for Boolean states (on, off), conditions (if, then, not), and conjunctions (and, or). Amino acid and nucleotide sequences are underlined. Boundaries are denoted by slash marks (e.g., “A/P”) and axes by hyphens (e.g., “A-P”). Short gene names (≤5 letters) are not usually abbreviated. Abbreviations include a.a. (amino acid), AEL (after egg laying), AP (after pupariation) a.k.a. (also known as), b.p. (base pair), h (hour), hs (heat shock), kb (kilobase), kD (kiloDalton), MC (macrochaete), mC (microchaete), St. (stage of embryogenesis), t.s. (temperaturesensitive), pers. comm. (personal communication), and unpub. obs. (unpublished observations). Times (h AEL or h AP) refer to a culture temperature of 25◦ C, unless stated otherwise. Polypeptide lengths are for the unprocessed (nascent) precursor. Genes that are usually called “neurogenic” (based on mutant phenotype) [436] are here termed “antineural” (based on function) [4387] to contrast them with “proneural” (based on function) genes [2018]. “Eye disc” refers to both the eye and antennal parts, and “wing disc” denotes the entire dorsal mesothoracic disc (wing, notal, and pleural parts). By tradition (quirky though it may be), fate maps employ left legs (Ch. 5), right wings (Ch. 6), and left eyes (Ch. 7) [185, 320, 526, 531], although right eyes are used by some authors [2962]. Readers must be familiar with the basics of ﬂy development [358, 2434, 3517] and the methods of modern genetics [354, 4671], including (1) induction of cell clones by ﬂp-mediated recombination [1530--1532, 3952, 4781] and the ﬂp-out trick [4159], (2) regional misexpression of genes via Gal4-UAS constructs [435, 3857], (3) temporal misexpression via heat-sensitive alleles [4214] or heat-shock promoters [2953], (4) enhancer trapping using lacZ reporter genes [278, 329, 1286, 4687], and (5) two-hybrid screening for protein interactions [222, 763, 1228, 1229, 1316]. Wherever possible, circuits are formulated in terms of Boolean logic [399] because this format shows syntax better than the “spaghetti diagrams” of genetics, electronics, or neural networks [2870]. The temptation to compare ﬂy circuits with vertebrate or nematode circuits

xiii

is generally resisted here for the sake of conciseness. Such comparisons can be found in Eric Davidson’s book Genomic Regulatory Systems [968] and at Tom Brody’s website The Interactive Fly. The term “link” is used in the sense of “causal linkage.” Links are symbolized as “ ” (activation) or “ ” (inhibition). When a gene is the object (e.g., “Dpp omb”), the effect is always at the transcriptional level, but pathways may be distilled in terms of either genes (en ci ptc) or proteins (En Ci Ptc), and any attendant ambiguities will be clariﬁed by context. Epistatic links need not be direct. Thus, “a c” could reﬂect a longer chain such as “a b c” or “a b c.” The reason for listing so many links in this book is to facilitate Aristotle’s goal of delineating the entire chain of causes from the fertilized egg to the adult [1993, 2919, 4305]. Only by concatenating all the known fragments can we see the gaps that remain to be ﬁlled. The terms “LOF” (Loss of Function) and “GOF” (Gain of Function) typically denote decreases or increases in levels of gene activity (i.e., under- or overexpression) [1117, 1455], but in the broader sense that will be used here, GOF also includes ectopic misexpression where the “gain” is regional (cf. Fig. 6.13). For example, clones of cells that express a wild-type allele of engrailed (en+ ) outside the territory where en+ is normally transcribed will be called “enGOF ” [4848]. Cases do arise where overexpressing a wild-type allele has effects that differ from expressing a constitutively active construct [3545], and these will be so indicated. Mutations that are neither LOF nor GOF (e.g., neomorphs and antimorphs) are rarer, and allele-speciﬁc superscripts will be retained for them (e.g., ci D [3818] and en1 [1636]). LOF and GOF tests are used to assess the necessity (LOF) and sufﬁciency (GOF) of a speciﬁc gene for a particular process [173, 3643, 4333, 4671], and they are valuable tools. However, neither is foolproof. For example, if we delete gene “a” and see no effect on bristles (a negative LOF result), then a is clearly dispensable for bristle formation, but we cannot conclude that a is irrelevant because it might be acting redundantly with gene “b” [2845, 4584]: “a or b bristle.” GOF data can also be misleading [6, 682, 1329]. For instance, if we drive the expression of gene “a” in a region where it is not normally transcribed and ﬁnd that it induces bristles (a positive GOF result), then a is clearly sufﬁcient for evoking bristles [1458, 1854, 2019, 3267], but this does not mean that a promotes bristle formation in wild-type ﬂies because GOF perturbations can saturate limiting components (e.g., bHLH

xiv

PREFACE

partners [438, 918, 1854] or external ligands [421]) or provoke interactions with other pathways (e.g., converging RTK cascades [326, 1117, 2623] or branched Frizzled chains [3912, 4365, 4867]), resulting in all sorts of artifacts [6]. Researchers beware!

the subtleties. Around every “gospel truth” there is a Talmudic aura of uncertainty. The author’s goal should be to make the material as accessible as possible without hiding any ambiguities. This book will attempt to do just that.

It is . . . unsafe to deduce normal gene function [when] the product is forced into inappropriate cells, perhaps in the absence of proteins with which it normally interacts and the presence of others that it does not normally encounter. [1304]

Nature is a home handywoman. Constrained by evolution, she does the job with the tools at hand, using a screwdriver for a hammer if necessary. . . . This machinery is neither elegant nor simple, but consists rather of a complex set of interacting proteins that were cobbled together by evolution. . . . Models help to organize our thoughts and offer testable hypotheses. Of course, in constructing a model, some data may need to be hammered into place, and the inconvenient data that cannot be coaxed into place have to be left out. The models that are frequently illustrated in minireviews . . . thus cannot be viewed as the ‘‘truth,” or they would narrow thought processes and squelch novel lines of research. We must be thoughtful iconoclasts, remembering that ultimately all models are wrong, fundamentally flawed or lacking the full complexity of systems shaped by evolution rather than intelligent design. We will thus use this forum to critique rather than prop up our model. It is increasingly clear that life is more complicated than portrayed there. [894]

Results derived from mutant analyses or from utilizing ectopic expression of a gene product reveal the potential of a particular interaction to occur, not whether the interaction actually occurs during wild-type development. [3248]

Artifacts can be minimized by combining LOF and GOF tests [147, 3462]. Indeed, that is the only way to distinguish factors that are “instructive” for cell fates from those that are merely “permissive.” Instructive agents have both LOF and GOF effects, whereas permissive agents have a LOF but no GOF effect [449, 1455]. Even this 2-pronged approach may not be able to resolve epistatic relations, however, where (1) interactions are cooperative as in multiprotein complexes, (2) pathways are nonlinear, (3) feedback obscures causality, or (4) the “upstream” vs. “downstream” ranking of genes contradicts the order of cellular actions in time. An example of the last difﬁculty involves scute and Notch. In general, scute is epistatic to Notch (i.e., scuteLOF NotchLOF ﬂies show the scuteLOF missing-bristle trait instead of the NotchLOF extra-bristle trait) [918, 1797, 1802, 3270, 3983], so scute should be acting downstream of Notch, but in fact scute must endow cells with “proneural competence” before Notch can enforce any “lateral inhibition” (cf. Ch. 3). The situation is even more complex at certain sites where Notch also acts before scute during a “prepattern” (pre-proneural) stage [461, 886]. Not all the ﬂy’s circuitry is as inscrutable as the Notchscute-Notch cascade, but our view of every subsystem is distorted by the imperfect lens of genetic dissection [2917, 3881, 4085, 4671]. Conclusions must therefore be qualiﬁed by layers of caveats about this or that alternative interpretation. The problem with such equivocation, of course, is that it can put readers to sleep. How much of this blather can readers tolerate? Why not just present “best guess” models and avoid all the dithering? Good advice on this issue comes from a delightful little essay entitled, “Wingless signaling: The inconvenient complexities of life.” Therein, Rachel Cox and Mark Peifer argue that cartoon-like abstractions are essential but must be tempered by critiques that convey

Only by venturing into the ocean of literature can novices experience the richer Fly World beyond the Internet harbors. Alas, it is all too easy to get lost in those rougher seas. For that reason, an effort is made to supply the equivalents of charts and buoys. To wit, all key mysteries that have taunted investigators are set in boldface when introduced. So are the models and metaphors that have been contrived to explain the mysteries, plus the epiphanies encountered whenever great mysteries were slain. All these concepts are inventoried in Appendix 2. Some of the coined names for the concepts are whimsical, but no more so than the silly names of many ﬂy genes. Indeed, working in this ﬁeld has been so much fun because of its playful irreverence – a legacy of the neophyte pioneers in Morgan’s team [119]. Even “the Boss” himself loved to clash ideas [2947] and smash idols [2946]. Ideas are contrasted here wherever possible, and the style is decidedly iconoclastic. All statements are source-referenced, and crossreferences that are not direct attributions are listed as “cf. such-and-such” – a style that is common in the humanities but rare in the sciences [1630]. The cf.’s mean to compare, confer, or just “see also.” Due to space limitations, some citation strings had to be truncated. Those cases are ﬂagged with a “” superscript to alert readers who want to trace earlier sources thereby. An unabridged bibliography is posted at The Interactive Fly.

PREFACE

Esoterica are banished to tables, ﬁgures, and appendices wherever possible, and supportive evidence is crammed into indented blocks of text so that readers can skip them if they want. Even so, readers may ﬁnd some sections of the text unnavigable without looking up the cited papers and tracing their lines of reasoning. Subheadings are worded as sentences so that the Table of Contents reads like a summary for each chapter. Gene abbreviations are deﬁned wherever they are used in the text. Overall, the layout is designed to avoid boring the expert without confusing the novice. I still remember how hard it was to make my way into this ﬁeld as an apprehensive apprentice. This ﬁeld has seen paradigm clashes of Promethean proportions, and those wars must be recounted to do the subject justice. For that reason, the modern facts have been woven into a historical tapestry, with a few homilies stitched in for good measure. Admitting past mistakes can help in spotting future pitfalls . . . even in the Olympian realm of molecular genetics, which surprisingly has more than a fair share of mortal foibles [1879, 2414, 3909, 4669, 4673]. The potential pitfalls include not only (1) the aforementioned LOF and GOF artifacts, but also (2) reporter anomalies (e.g., perdurance of β-gal [3764, 4188]), (3) antibody limitations (e.g., misleading epitopes on proteins that are cleaved [155, 3271] or reshaped [1980]), (4) confocal illusions [3293, 4760], and (5) in vitro inﬁdelities relative to in vivo conditions [655, 871]. For the next generation of researchers, some of the parables may sound quaint, but for those of us who toiled through this period, they are a chronicle worth preserving. Readers accustomed to color photos may bemoan the book’s reliance on black-and-white diagrams. I am sorry for any disappointment. The latter style just seemed more ﬁtting for an abstract analysis. All the ﬁgures were drawn in adobe Illustrator by me (a hopeless attempt to compete with my truly artistic siblings). They evolved from cartoons into montages. When many grew too big to ﬁt the standard 6 × 9-in. size of this series, I tried breaking them into pieces but found that the surgery was lethal. The montages had acquired a life of their own. They tell whole stories (some of which spill over into App. 7). I thank Cambridge for approving a

xv

larger trim size and for letting me set my own deadline. The cusp of the millennium seemed an apt time to step back and take a wide-angle “snapshot” of this blossoming ﬁeld. The last batch of citations came from the annual Drosophila Research Conference (in Washington, DC) entitled, “2001: A Fly Odyssey.” This project began in 1992 when Robin Smith (then Life Sciences Editor at Cambridge) asked me to write a book for this series at the behest of Paul Green (a series editor). The topic took shape gradually, and the contract was signed in 1996. By 1997, my other professional pursuits had to be sidelined as the writing became allconsuming. I thank Peter Barlow (another series editor) for calming my fears and Ellen Carlin (Assistant Life Sciences Editor) for trusting my judgment. Encouragement was provided by my dear parents (Maj. Gen. Lewis I. Held and Minnie Cansino Held), siblings (Lloyd, a.k.a Grey, and Linda), other relatives and sundry friends – most of whom remain skeptical that any sane adult can adore ﬂies. Maybe this book will change their minds? Probably not! Critical comments on portions of the manuscript were kindly furnished by Seth Blair, Tom Brody, Ian Duncan, Matt Gibson, Robert Holmgren, Teresa Orenic, Grace Panganiban, Amy Ralston, Allen Shearn, David Sutherland, and Tanya Wolff. The idea about Notch and Argos in the Skeptic-Theorist debate (Ch. 6) was Seth’s. I regret any overlooked errors. As one foot soldier in the global army of ﬂy pushers, I have met many “generals” over the years who ﬁgure prominently in this saga. By far the greatest – and humblest – was Curt Stern. His musings on the mysteries of patterning were the siren songs that lured me to this lovely ﬂy. Those of us who heeded his call have long dreamt of ﬁnding insights one day. Little did any of us suspect, though, that the bounty of revelations in the last decade would go so far beyond merely sating our curiosity. As we sift the treasure, the sparkle of so many answers is fostering – even in the saltiest among us – a profound sense of awe. Lewis I. Held, Jr. Lubbock, Texas April 2001

CHAPTER ONE

Cell Lineage vs. Intercellular Signaling

Imaginal discs are hollow sacs of cells that make adult structures during metamorphosis. They are so named because “imago” is the old term for an adult insect [4008], and their shape is discoid (i.e., ﬂat and round like a deﬂated balloon) [377]. They arise as pockets in the embryonic ectoderm and grow inside the body cavity until the larva becomes a pupa, at which point they turn inside out (“evaginate”) to form the body wall and appendages [3165]. In a D. melanogaster larva there are 19 discs (Fig. 1.1). Nine pairs form the head and thorax, and a medial disc forms the genitalia. The abdominal epidermis comes from separate cell clusters called “histoblast nests” [2301, 2648, 3647]. Unlike discs, histoblast nests remain superﬁcial during larval life [927] and do not grow until the pupal stage [2650]. Given the diversity of cell types in the adult skin (e.g., bristles, sensilla, photoreceptors) and the commonality of their descent from one progenitor (the fertilized egg), it is natural to ask how cells specialize to adopt divergent roles. In principle, cells can acquire instructions from ancestors or contemporaries [1654]. More speciﬁcally, a cell can inherit predispositions from its mother (“intrinsic” mode), take cues from neighbors (“extrinsic” mode), or both [477, 1614, 2019, 2451, 3741]. The predispositions could be gene states, while the cues could be diffusible ligands [1144, 3182]. To the extent that fates are assigned intrinsically, there should be a rigid correspondence between (1) parts of the anatomy and (2) branches of the lineage tree [1362, 4086, 4087]. That is, a clone of cells descended from ancestral cell “x” should make structure “X”, while another clone descended from ancestor cell “y” should make structure “Y”. Moreover, these rules should be obeyed

in every member of the species. C. elegans worms adhere closely to this strategy [1284, 4201, 4202], but ﬂies do not [1839, 1881]. In D. melanogaster, the only adult structures that use an intrinsic mode are tiny sense organs [532, 1410, 3441]. All larger parts of the body use extrinsic mechanisms. Thus, the problem of how discs develop can be reduced to questions about how cells communicate [695]. Who signals to whom? Over what distance? With what molecules? To what end?

Discs are not clones Proof that cell pedigrees are irrelevant for disc patterning was ﬁrst provided in a 1929 paper [4180] by Alfred Henry Sturtevant (1891–1970) – a wunderkind of the Morgan lab [257, 2504, 2615]. Sturtevant studied a strain that produced freakish ﬂies called “gynandromorphs” [2950]. Each such ﬂy is a patchwork of purely male and female tissues (Fig. 1.2) [1715]. They begin life with two X chromosomes but typically lose an X during the ﬁrst mitosis, so that one of the two zygotic nuclei becomes 1X [1695]. Because gender in ﬂies is dictated by the number of Xs relative to the numbers of autosomes [817], the 1X nucleus – and the half of the body that it populates – becomes male. Sexual traits are expressed autonomously at a single-cell level because ﬂies lack circulating sex hormones. The male/female boundary can be mapped throughout the cuticle (not just in dimorphic organs) by using recessive mutations to mark one of the Xs. The yellowLOF mutation is often used because it turns the normally brown bristles (and cuticle) yellow [4101]. Such ﬂies are useful for cell lineage analysis because any body part that develops clonally must come 1

2

IMAGINAL DISCS

3 1

2 5

4

8

9

Discs

h 7 6

10

1 2 3 4 5 6 7 8 9 10

Labial Clypeolabral Humeral Eye-antennal 1st Leg 2nd Leg 3rd Leg Wing Haltere Genital

FIGURE 1.1. Imaginal discs and their cuticular products. The ﬂy exterior is assembled from separate parts (like an automobile).

The epidermis of the head and thorax come from 9 bilateral pairs of discs (one of each kind is shown), and genitalia come from a medial disc, so there are 19 discs total. Abdominal wall comes from histoblast nests (h): tergites from dorsal nests, and sternites and pleurae from ventral nests. Discs are drawn to the same scale, and are oriented to display their mature shapes and folding. Placements are approximate. Clypeolabral and labial discs are attached to the pharyngeal skeleton (black hooks) [3285], while other discs adhere to other larval organs (not shown) [527,834,4565]. “Humeral” is synonymous with “dorsal prothoracic” disc. Bristles are omitted for clarity, and ﬂank sclerites are simpliﬁed. An adult fruit ﬂy is ∼3 mm long. Full-grown larvae are roughly twice that length [3421]. About half the larval midsection is omitted here. Adapted from [1739,4565]. Discs look more alike than the structures they produce. The same is true at the cell level, where discs are nearly indistinguishable by ordinary histology [3165,4424]. Even at the molecular level, different discs make virtually identical suites of proteins [1459,1611,3625,3756,3865], although amounts vary. The reason for these common features – as later chapters show – is that all discs use the same basic “toolkit” of molecules for intercellular signaling [662], although the circuitry (i.e., how those molecules interact) is tailored to the disc-speciﬁc patterns [1440].

from one single male or female progenitor cell and hence be purely yellow or brown. Sturtevant discovered that cuticular derivatives of all the larger discs can be bisected by a yellow/brown

boundary. Hence, these discs do not develop as clones. Subsequent studies found mosaicism in the smaller discs as well [1370, 2026, 2029, 2828]. By implication, each disc must come from ≥2 cells [2411]. In fact, when discs are ﬁrst

CHAPTER ONE. CELL LINEAGE VS. INTERCELLULAR SIGNALING

P

3

A

A y LOF y LOF

D

y+

L

P

X D

X X

R L

R V

V

isc ng d wi

D

R

wing dis c

isc ng d wi

L

D

R

wing dis c

L

V

V

fly #1

fly #2

FIGURE 1.2. The nonclonal nature of ﬂy development. The irrelevance of cell clones to pattern formation is seen in the piebald

variegation of sexually mosaic “gynandromorphs” (middle panel) [1370,2026]. Such ﬂies are typically half male (gray) and half female (black) [1715,2950]. They start life as a heterozygous female (2X) zygote but lose an X chromosome from one nucleus at the ﬁrst mitosis to create a male (1X) clone (ﬂy 1, top panel) [1695]. If the X that remains has the yellow LOF (y LOF ) allele (enlarged gray circle), then the descendants of that nucleus will make yellow (instead of brown = wild-type) bristles or cuticle in the adult (ﬂy 1, bottom panel). The two embryos at the top of the ﬁgure (A, anterior; P, posterior) differ in the orientation of the ﬁrst mitotic spindle [3274,4021]. This disparity causes the male/female boundary to trace different paths in the cuticle (middle panel) [4649,4652,4845]. The adults are bisected in the middle panel, and the cross-sections are turned ∼90◦ to a frontal view in the bottom panel (D, dorsal; V, ventral; R, right; L, left). The outer ring of circles (nuclei) schematically represents the thoracic epidermis. The inscribed “tree” represents an imaginary series of mitoses (branch points) from the initial two nuclei to the adult epidermis. Bristle numbers and cell densities are drastically reduced for clarity. If the wing disc (dashed outline) were a clone – i.e., derived from a single nucleus – then it should be purely yellow or brown because its progenitor nucleus must be one or the other. In actual gynandromorphs, however, the wing disc is often mosaic (R disc in ﬂy 1 and L disc in ﬂy 2), so it cannot be a clone. Moreover, the ability of the male/female boundaries to pass between any two landmarks (e.g., the different pairs of bristles in ﬂy 1 vs. ﬂy 2) argues that the patterns of cell lineage within the disc (inscribed trees) must also vary from ﬂy to ﬂy. Overall, therefore, such ﬂies reveal a fundamental uncoupling between pedigrees and patterning. This uncoupling is abstractly seen in the ability of the male/female “hour hands” to lie anywhere on the epidermal “clockface.” The two ﬂies shown here are only two examples from a large set of possibilities.

4

detectable histologically, each contains at least 10 cells (cf. Ch. 4). It is a quirk of history that the full import of Sturtevant’s study was only realized 40 years later [119] when Antonio Garc´ıa-Bellido and John Merriam used Sturtevant’s data to map the embryonic disc primordia [1370].

No part of a disc is a clone, except claws and tiny sense organs Yellow/brown gynandromorphs are as eye-catching as a herd of Appaloosa horses because each individual has a unique pattern of colored patches (Fig. 1.2) [2026]. Their harlequin variegation is due to (1) the random orientation of the zygote’s ﬁrst mitotic division in all three dimensions from one individual to the next [3274, 4021] and (2) the tendency of sister nuclei to stay together during cleavage [4899]. The male/female line hence intersects the egg surface at random angles [4222, 4845], and the yellow/ brown boundary should bisect any given area of the adult surface if a sufﬁciently large population is examined – unless that area is delimited clonally. Among the 96 specimens that Sturtevant analyzed, many groups of cuticular landmarks were divided by such boundaries. This “indeterminate” cell lineage was epitomized by two pairs of bristles that belong to the wing disc: the dorsocentrals and postalars. From one ﬂy to the next, Sturtevant found that both dorsocentrals may be alike [i.e. both male or both female] but different from both postalars, or the posterior dorsocentral and posterior postalar may be alike but different from the corresponding anteriors, or any one of the four may be different from the other three. Such relations occur for any group of mesonotal bristles one examines. [4180]

IMAGINAL DISCS

Indeed, male/female boundaries meander relatively freely through every bristle array on the adult surface [1800, 2026, 3007, 3539, 4652]. Clearly, discs are not balkanized into subregions where individual cells obey commands such as “Divide ‘n’ times and tell your descendants to make this part of the adult.” The only exceptions are (1) bristles and sensilla [3441] whose few component cells (≤10) come from single “mother” cells and (2) claws [1356], which follow a similar developmental path [1587]. Additional instances are found in embryonic development – e.g., neural ganglia [627], muscle subtypes [250, 3684, 3698], and cardiac precursors [1339, 4194, 4547]. Wherever cell-type determination is uncoupled from cell lineage – as here in the case of large-scale patterning within discs – it must perforce rely on intercellular signaling [293, 354, 4727].

Cells belong to lineage ‘‘compartments’’ Despite the rarity of rigid pedigrees in disc development, cells commonly obey looser edicts such as “You may make any portion of region ‘R’, but nothing outside it” [4671]. Regional limits of this kind were discovered in wing discs when marked cells were spurred to grow faster than background cells. Oversize anterior or posterior clones grew up to – but failed to cross – a boundary that roughly bisects the disc [1376, 1377], and analogous “compartments” were later found in halteres [1358, 1771], legs [1800, 2449, 4076], antennae [2931], genitalia [1107, 2028], and the proboscis [4144, 4145]. Compartments are essential for patterning (cf. Ch. 4 ff ), but their lineage constraints per se are not [754, 2428, 2448, 2677, 4491]. Hence, the existence of these clans does not negate the “Proximity vs. Pedigree Rule” [3445] enunciated above. Put simply, this rule asserts that cells select fates based on input from peers, not parents [354, 526, 1808].

CHAPTER TWO

The Bristle

Tactile stimuli are hard for arthropods to detect through the armor of their rigid exoskeleton [1666, 3582]. To solve this problem, ﬂies use bristles (Fig. 2.1). When a bristle is deﬂected, the pivoting of the shaft in its socket deforms the dendrite of a neuron attached to the shaft’s base [789, 1352, 2174, 2787]. The resulting depolarization sends an action potential to the central nervous system (CNS) [1118, 2173, 2196, 4527]. Flies can pinpoint sensations because axons from different bristles get “wired” to different CNS target cells during metamorphosis, although much remains to be learned about the topology of these neurosensory maps (cf. Ch. 6). Mechanosensory bristles are formed by 5 cells: 2 superﬁcial cells that secrete cuticle (the shaft and socket cells) and 3 subepidermal cells that do not (the neuron, sheath, and glial cells) [2475, 3351, 3552, 3832, 4531]. These 5 cells descend from a “sensory organ precursor” (SOP). The SOP divides to produce one daughter (IIa) that yields the outer cells, and another (IIb) that yields the inner cells [1447, 1741, 1925]. The sheath cell wraps the neuron’s dendrite [602, 789, 3351], while the glial cell wraps the axon [2173]. A sixth cell – the “bract cell” – is found in association with bristles on the distal leg and proximal wing [524, 1714, 1808]. It secretes a thickened hair (“bract”) that is pigmented like the bristle shaft but much smaller [3362, 3421]. The bract cell is not part of the SOP clone [1808]. The way in which it is recruited from epidermal cells is discussed later. Until 1999, the glial cell’s origin was obscure [1463, 1465, 1741, 1925], and only the shaft, socket, sheath, and neuron were considered to comprise the SOP clone. In 1999, a debate about the sequence of bristle cell mitoses [2680, 3550] prompted a reinvestigation of the mitoses themselves [1447, 3549], whereupon a new mitosis was

discovered. It had hitherto been overlooked because the glial cell is small and migrates away from its birthplace. Pre-1999 models are being revised to include this amendment [2382]. Chemosensory bristles have all the components of a mechanosensory bristle plus 4 extra neurons, whose dendrites project to a pore at the shaft’s tip [1741, 3061, 3529, 4841] where they detect chemicals (Fig. 2.8) [3835]. Strangely, such bristles (on the legs and wings at least) are also photosensitive, with independently entrainable circadian clocks [2333, 3327, 3401]. Aside from sensory modality [3005], ﬂy bristles also vary in size, shape, pigmentation, and pattern. Bristles are intriguing not only because their stereotyped mitoses violate the general rule of indeterminate lineage (cf. Ch. 1), but also because they encapsulate the problem of differentiation (how do cells acquire differences?) [2424, 2577, 4658]. In theory, the instructions for assigning fates could be unequally inherited from the SOP, with no need for cross-talk among descendants. According to this “Obey Your Mother! Model,” bristle cells adopt fates based on cues inherited from their mothers. The main cue appears to be the presence or absence of a membrane-associated protein called “Numb.” Numb has all the features expected for a heritable determinant of cell fate.

Numb segregates asymmetrically and dictates bristle cell fates The gene numb was isolated in a screen for mutations affecting the embryonic peripheral nervous system (PNS) [4417]. In a seminal 1994 article that provided the key to deciphering bristle differentiation, Michelle Rhyu et al. 5

6

IMAGINAL DISCS

SOP apical

P

A s al

mod subp ular rogr am

ba

Asymmetry

Numb crescent 0:00

Mitosis

P

A

Memory?

0

Reset

1

IIa 0

1 IIb o

90 rotation of spindle

1 2:44

0 0 0

l

10

5:00 l

10 10

101 cell

Neur o

n

ll th ce Shea

t cell Shaf

Sock

et ce

ll

100

basa

01

apica

00

cell

A l

P

11 Glial

IIIb 10

3:40

Glial

Bract cell

?

basa

l apica

Bract

1 1

Differentiation

Cuticle

Epidermal cells

To CNS

CHAPTER TWO. THE BRISTLE

in San Francisco reported that numb mutations also affect adult bristles [3579] and, more important, that Numb protein is distributed unequally during SOP divisions. Indeed, this was the ﬁrst gene product in ﬂies ever shown to segregate asymmetrically in mitosis, although others soon followed [2021]. Within the SOP lineage, 4 cells inherit Numb (IIb, shaft cell, glial cell, neuron; Fig. 2.1) [1447, 3579, 4542] while 4 do not (IIa, socket cell, IIIb, sheath cell), and mutant defects are generally consistent with this parceling. Thus, numb null mutations cause SOPs to produce 4 outer cells and no inner ones – implying that IIb adopts a IIa fate – and the outer cells are often all sockets, so a shaft-to-socket transformation must also be involved. A third conversion (neuron-to-sheath) occurs in hypomorphs [4542]. Collectively, these phenotypes imply a fate-assigning role for numb at every mitosis in the lineage, with the possible exception of the glia-producing IIb mitosis, which, as mentioned above, has only recently begun to be studied. The history of a cell’s Numb states can be denoted by the left-to-right order of digits in a binary code (Fig. 2.1), where “1” signiﬁes Numb’s presence and “0” its absence. Thus, the various sister cells in the SOP lineage would have the following paired codes: IIa (0) vs. IIb (1). Socket (00) vs. shaft (01). IIIb (10) vs. glial cell (11). Sheath (100) vs. neuron (101). From the standpoint of a strict “Coding Model,” the code would be causal. That is, a bristle cell’s fate would

7

be dictated by the series of Numb states (0 or 1) experienced by its ancestors. This code would explain the null phenotype where all cells assume a 00 (socket) state, and it would also explain the hypomorphic condition where neurons (101) switch to sheaths (100). To wit, leaky Numb levels might be high enough to let IIb attain its “1” state but not to push neurons into their later “1” state. One test of this model would be to overexpress numb. Flooding the lineage with Numb protein should raise all “0” states to “1” and cause all cells to differentiate as glia (11). When UAS-numb is driven by a Gal4 transgene expressed in SOPs, no clusters of 4 glial cells were reported [4542]. The most extreme defect was a 4-neuron trait where IIa likely became IIb (0 1) and sheath cells became neurons (100 101). Milder abnormalities were also seen, including “2 sheaths: 2 neurons” (0 1 but not 100 101) and duplicated shafts (00 01 but not 0 1). Overall, the data agree with the model, although the failure to force cells into a glial fate is problematic. Perhaps the excess Numb cannot prevent Numb’s level from being reset to “0” in IIb (Fig. 2.1). Additional support for the model comes from ﬂies carrying a hs-numb construct (numb joined to a heatshock promoter). When such ﬂies are heat-shocked around the time of SOP mitoses, they display “2 sheaths: 2 neurons” as well as “2 shafts, sheath, neuron” (socketto-shaft conversion) and “socket, shaft, 2 neurons” (sheath-to-neuron). These defects are explicable by the forced presence of Numb in the IIa (0 1), socket (00 01), or sheath (100 101) cell [3579]. Four-neuron

FIGURE 2.1. Development of a mechanosensory bristle from a sensory organ precursor (SOP). Compass (upper left) gives initial

directions (A, anterior; P, posterior). Times (hours: minutes at 23◦ C) are for microchaete mitoses on the notum but are similar for other bristles [1447]. The SOP arises from an ordinary epithelial cell. It starts to divide (at ∼16 h after pupariation) to form IIa and IIb. IIa’s daughters will make a socket and shaft. IIb’s daughters are IIIb and a glial cell. The glial cell is smaller and buds off basally in the manner of a CNS neuroblast division [1073, 1740]. IIIb divides to form a sheath cell and neuron. Some bristles have a thick hair (“bract”) atop their sockets (inset), which is made by a clonally unrelated cell. Each mitosis obeys stereotyped steps (dashed trapezoid) that comprise a modular subprogram: (1) Numb localizes to one side of the cell cortex (crescent), (2) segregates to one daughter, and (3) alters cell fate. Letting 1 and 0 signify Numb’s presence or absence, each cell can acquire a unique code if it “remembers” its former Numb states. Imaginary memory registers (underlined spaces) are shown for a few cells, with left-to-right order recording successively later states. For such a binary code to work, IIb must eliminate (“reset”) Numb before dividing. When SOPs are prevented from dividing, they become neurons [1743]. This result has been interpreted as a default condition, but it may instead reﬂect persistence of Numb: the continual presence of Numb should lead to a “nonsense” code (111) that might be interpreted as “neuron” (101). The mechanism whereby cells remember former Numb states is unknown. Timing and branching of the pedigree are as per [1447, 3549]. Other details are based on [1449, 1741, 1808, 3579]. See [3195] for lineage comparisons with other sensory organs. N.B.: Grooves are absent from some bristles (e.g., sex comb teeth [1714]). Epidermal cells are sometimes aligned with this degree of precision [2388], although they need not be. Chemosensory bristles have 4 additional neurons (cf. Fig. 2.8) [4125], and their SOPs obey a different lineage [3529]. See also App. 7.

8

IMAGINAL DISCS

?

1cleave

cleave

P

P

* Notch* Not ch

Delta

?

Su(H) Delta

2 move

P

*

Su(H)

can't move because tethered by Numb

P

3

P Su(H)

ch

* Notch* Not ch

Numb

ON

OFF

E(spl)

Nuclear Notch Model

E(spl)

Socket

Shaft

Catalysis Model ?

? Su(H)

Su(H) P

* Notch*

P

Notch Delta

* Notch* Notch

Delta

1 activate

P

P

*

Su(H)

Numb

can't activate because blocked by Numb

2

ON

OFF

E(spl)

E(spl)

Socket

Shaft

CHAPTER TWO. THE BRISTLE

phenotypes were not observed, probably because pulses were too short to affect all three rounds of (asynchronous) mitoses. If Numb were a traditional “cytoplasmic determinant,” then it would specify only one type of tissue or cell [1904]. On the contrary, it marks 4 different cells in the SOP lineage. Moreover, it plays similar roles in sense organs of the larval PNS [3579, 4417], in neuroblasts of the embryonic CNS [2451, 3579, 4028, 4523], in cardiac cell progenitors [1339, 4194, 4547], in sibling founder cells of larval vs. adult muscles [653, 3684], and in muscle subtype determination [251, 910, 3263, 3687]. Thus, its role transcends histotype. Evidently, Numb functions as a versatile switch that enables daughter cells to become different from one another, regardless of what those differences may be [831, 1761, 3263, 3579, 4875]. As a binary digit (“bit”), Numb is the best example ever adduced that ﬂies can use abstract symbols for instructions just as computers employ machine language. As explained below, this “Numb Epiphany” of 1994 is not only helping to elucidate how genes can work as switches, but it is also revealing how an intrinsic mechanism of fate speciﬁcation can dovetail with an extrinsic pathway of intercellular signaling.

Delta needs to activate Notch, but not as a signal per se Although the Numb code should be sufﬁcient for assigning all fates, some cell interactions have also been implicated. The 4-neuron trait that is caused by gain-offunction (GOF) numb manipulations is also seen with loss-of-function (LOF) mutations in Delta (Dl) and Notch (N) [1742, 3272]. Because Dl and N mediate “sibling rivalries,” whereby equivalent cells become different (cf. Fig. 3.6) [2222, 3022], they could – in theory – create binary codes by refereeing a series of bouts (winner = 1; loser = 0) without relying on cell pedigrees at all [1614].

9

Might fates be computed by either lineage (via Numb) or signaling (via Dl and N), with one agent assuring success if the other fails? No, because such redundancy would imply that phenotypes should be wild-type unless both strategies fail, but (as stated above) fates can be altered by single LOF mutations in numb, Dl, or N. Rather, it seems that the two devices are connected in series, not in parallel. Dl and N are transmembrane proteins that interact as ligand (signal) and receptor (receiver), respectively [1204, 2626]. When a N-expressing cell contacts a Dl-expressing cell, N is activated by dimerization [3022] or oligomerization [2209, 2299]. Activation causes N’s intracellular domain (“N-intra”) to detach from the membrane and go to the nucleus, where it stimulates transcription of target genes [4155]. Numb may block signaling by tethering N-intra to the cortex (Fig. 2.2), thus keeping it from reaching its targets. Enough Numb would normally be present to sequester all N, although an artiﬁcial excess of N could escape Numb’s grasp and cause the kinds of N GOF phenotypes that are seen [1307, 1651]. The need for ligand may suggest extrinsic signals, but there cannot be any instructive (on/off) signaling per se [1433] because Dl is expressed at equal levels in IIa, IIb, and surrounding nonbristle cells [3270]. Evidently, Dl plays only a permissive role, essentially like a seaman sending Morse code by using a shutter (Numb) to blink a light (Dl-N) that stays on. This “Blinker Model” supposes that Dl’s job is merely to keep N active so that the nucleus only gets a “N = off” signal when Numb is present. Mosaic analyses suggest that the SOP descendants themselves supply one another with the ligands for N stimulation, with no reliance on surrounding epidermal cells [4859]. This intrabristle cross-talk has been conﬁrmed in an interesting experiment. When Dl is overexpressed in the neuron, the adjacent shaft cell

FIGURE 2.2. Models for Notch signaling and its blockage by Numb. Black rectangles are proteins, and connecting “wires” are

binding sites. Contact with Delta ligand on a neighbor’s surface activates (asterisk) the Notch receptor, possibly by dimerization (partner outlined) [3022]. Cells that lack Numb (left) can relay the signal to its nucleus, while those that express Numb (right) cannot. The models differ in how Numb stops the signal. In the Nuclear Notch Model (above the line) [1307, 1448, 1651, 2299, 4027, 4244, 4542], Numb stops Notch from leaving its roost (ghost image) by anchoring it to the membrane [2267] via an unknown linker (“?” = possibly Partner of Numb [2609]). In the Catalysis Model (below the line) [112, 132, 1131, 3022, 4244], Numb blocks an active site for Su(H) activation (covalent modiﬁcation?). Numb is shown binding Notch at a phosphotyrosine (P), but Numb’s PTB domain is unusual and may not need a phosphate [2530, 4789], and Notch is only known to have phosphoserines [2209]. Notch resides in the apicolateral membrane [184, 1203, 1448, 2070]. The cell’s apex is carpeted with microvilli. Su(H) can activate transcription (right-angle arrow) of E(spl) (a.k.a. “m8”; cf. Fig. 2.4) by binding its promoter (gray rectangle), but E(spl) may not dictate bristle cell fates, nor is Su(H) needed for signal relay in neurons or sheath cells (see text). Estimates are that a signal at the membrane takes ∼20–90 min to cause detectable changes in target gene expression [184]. See also App. 7.

10

transforms into a socket cell (Fig. 2.7c) [2008]. Clearly, Numb’s lock on the Notch pathway can be artiﬁcially overridden by excess Dl. In contrast to the Blinker Model, the popular view has been that Numb merely biases Dl-N contests [1613, 2019, 2021, 2222, 3437], rather than being the sole deciding factor. Yuh Nung Jan and Lily Yeh Jan, who pioneered this ﬁeld, advocated this “Bias Model” but recognized an inherent paradox: because one sister cell should win every contest (with or without a Numb handicap) the numb null phenotype should be wild-type, but it is not (and the same dilemma applies in the CNS [552]). To explain why, they invoked time constraints [2020]: We think that . . . an intrinsic mechanism utilizing numb protein is superimposed on the Notch/Delta system to bias the competition. . . . We speculate that this Notch/Delta system is not sufficiently reliable to ensure that the two cells always acquire two different fates in the allotted time. (In the case of IIa vs. IIb fates, the time window is less than 2 hr.) . . . This hypothesis could explain the variable phenotype resulting from complete loss of numb function. In numb null clones some sensory bristles show the severe phenotype of having four socket cells, whereas other sensory bristles develop normally. Our interpretation is that, in the absence of numb, the Notch/Delta system still operates, but is not sufficiently reliable. . . . Some sensory bristle cells were able to finish the competition and form normal sensory bristles with four distinct fates, whereas others were unable to do so.

The Bias Model predicts that contests will end in Dlrich/Dl-poor (winner/loser) cell pairs. On the contrary, only Dl-equivalent pairs are detected in wild-type ﬂies [3270]. Rejecting the Bias Model in favor of the Numbdictated “Obey Your Mother!” Model still leaves the question of why all bristle sites in numb null clones do not have a 4-socket phenotype [2019]. Perhaps the normally dormant Dl-N rivalry mechanism has been awakened in these clones, in which case they should manifest Dl-rich/Dl-poor cell pairs (a testable prediction). Alternatively, unknown asymmetries may be augmenting Numb’s function (i.e., a partial redundancy). Either way, Numb’s control over N begs the evolutionary question: how did a heritable determinant (Numb) “hijack” an intercellular signaling pathway (Dl-N)?

Amnesic cells can use sequential gating to simulate a binary code If Numb is the bit in the bristle formula, then how do cells interpret 2- and 3-bit “words” for the various cell

IMAGINAL DISCS

types? A simple ratcheting mechanism, whereby cells count how many times they have been “1,” cannot sufﬁce because in that case “01” and “10” would be synonyms. It would seem that cells must use some sort of combinatorial code where genes aside from numb are used for recording previous Numb states. Figure 2.1 illustrates such a Coding Model. Do any known genes behave like a primary memory register – namely, their mutant alleles convert IIa into IIb (or vice versa) without switching any subsequent states? Among the genes whose mutant phenotypes connote a IIa-IIb switch, only Bearded (Brd) lacks later effects (App. 3) [2500]. Its GOF phenotype consists of neurons and sheath cells without shafts or sockets – indicative of a transformation of IIa into IIb. Thus, Brd could store the outcome of the ﬁrst mitosis. (N would turn Brd off in IIa.) Brd null mutants look wild-type [2500], but this impotence is attributable to redundant paralogs [2382]. There is another way of thinking about Numb’s mode of action that does not involve memory genes per se. To wit, Numb’s ﬁrst state (0 or 1) might simply “gate” IIa and IIb into divergent signal transduction pathways (STPs), so that the second Numb signal (0 or 1) is interpreted differently by IIa daughters (STP 1) vs. IIb daughters (STP 2). According to this “Gating Model,” genes that act only in the IIa STP should interconvert shafts and sockets when mutated, but should have no effect on neurons, sheath, or glial cells (IIb descendants), and separate sets of STP genes would operate exclusively in the IIb and IIIb sublineages. Indeed, Suppressor of Hairless appears to be a IIaspeciﬁc STP gene. Null Su(H) mutations suppress only part of the phenotype caused by numb LOF – namely, the shaft-to-socket switch but not the neuron-to-sheath switch – implying that Su(H) is only needed in the IIa lineage [4542]. This conclusion is bolstered by the ability of excess Su(H) to transform IIa (shaft-to-socket) but not IIIb daughters [200, 3827, 4542]. Su(H) is detected in both the IIa and IIIb lineages, but its level is highest in the socket cell (as are Su(H) transcripts [3826]) – a IIa daughter [1448]. Su(H) moves from the socket cell’s cytoplasm to its nucleus when N is activated [1448] – precisely the behavior expected for a messenger molecule [1269, 1307]. Su(H) can bind both to N (signal acquisition?) [1269] and to DNA sites (signal delivery?) upstream of genes in the Enhancer of split Complex [1131, 2453], which may control bristle cell fates (but see below). Thus, Su(H) has not only the phenotypic properties of a IIa STP agent, but also the histological hallmarks.

CHAPTER TWO. THE BRISTLE

When homozygous Su(H) null clones are induced near the time of SOP mitoses (3rd instar or pupal period) they transform sockets into shafts [3827, 4542], but when they are induced earlier (1st or 2nd instar) most sockets and shafts are missing and replaced by neurons and possibly sheath cells [3820]. This switching of IIa to IIb implies that Su(H) mediates not only the socket/shaft decision but also the IIa/IIb decision. If so, then the signal relays in STP 1 would be as follows, where States 0 and 1 indicate Numb’s absence or presence, “ ” and “ ” denote activation or inhibition, and “default” refers to the state of a cell in the absence of extrinsic signals: 1. Decision to become IIa vs. IIb (ﬁrst mitosis): State 0 (N = on): Dl N Su(H) {genes in Set 1} IIa. State 1 (N = off): Numb N. {Genes in Default Set 2} IIb. 2. Decision to become socket vs. shaft cell (second mitosis): State 0 (N = on): Dl N Su(H) {genes in Set 3} socket. State 1 (N = off): Numb N. {Genes in Default Set 4} shaft. Although it is easy to imagine how N could toggle cells from one set of genes to another, it is unclear why cells that do not receive a signal should switch their sets of default genes (Set 2 to Set 4) from one generation to the next. Conceivably, the latter is caused by a separate mitosis-counting mechanism. Counting of mitoses could be accomplished by a digital ratchet or an analog quantiﬁer [1880, 2762]. An example of an analog device exists in the ﬂy CNS, where descendants of the NB4-2 neuroblast apparently count mitoses of ganglion mother cells via the quantity of two homeodomain proteins (Pdm-1 and Pdm-2) whose concentration slowly falls [313, 314, 3713, 4804, 4816]. Alternatively, SOP progeny might get their timing cues from the cascade of genes that respond to ecdysone [247, 2111, 2134, 3059, 3814]. Indeed, SOP mitoses malfunction at speciﬁc macrochaete sites when ecdysone levels are depressed [3990], possibly due to this hormone’s management of disc cell cycles [1185, 1599]. It should be possible to distinguish intrinsic vs. extrinsic clocks by letting SOPs differentiate in vitro, as has been done for neuroblasts [485]. With a “schedule” strategy (i.e., events A, B, and C are triggered independently at clock times 1, 2, and 3), later steps should occur

11

when early ones are blocked; whereas with a “dominoes” strategy (A licenses B which licenses C), later steps should not proceed in the absence of earlier ones [1805, 2943, 3010]. In either case, the phenotype expected for LOF mutations in “counting genes” would be completion of SOP mitoses but no differentiation because descendants will wrongly think that they are still in the ﬁrst generation [916]. One gear in the escapement mechanism could be the p69 Tramtrack protein, which helps specify bristle cell fates (see below). Overexpression of Tramtrack during part of the cell cycle can halt bristle differentiation [3502], but it also blocks mitosis. Prospero has similar effects in the CNS [2522]. Intriguingly, LOF mutations in bazooka cause extra cell divisions in the SOP lineage [3629]. Bazooka is needed for neuroblast asymmetry and for adherens junctions [2985, 3803, 4709]. Other possible cogs in the clockwork include (1) Dacapo, a cyclin-dependent kinase inhibitor [6, 1002, 1004, 2404]; (2) Twins, the regulatory subunit of a mitosis affecting phosphatase (see below) [6, 1535, 2761]; and (3) String, a Cdc25 phosphatase that is rate-limiting for mitosis [6, 2740, 3081]. Interestingly, string has separate cis-enhancers for epidermal vs. SOP mitoses [2480]. In nematodes, some Notch-dependent fates are indeed gated by cell-cycle phases [79], and a few recently recovered ﬂy genes that are suspected of being SOP gating agents also interact with Notch [6]. Assuming that a gene in Set 2 diverts IIb into a new transduction pathway, a second messenger other than Su(H) would need to mediate the sheath/neuron decision. Su(H)-independent Notch pathways operate elsewhere [2353, 2453, 2540, 2549, 2747, 2839, 3694, 3695], so this idea is plausible [444, 4542, 4581]. In fact, a IIb-speciﬁc transducer called “Sanpodo” (Spdo) has been identiﬁed for the larval PNS, and it plays a similar role in the CNS [3985]: (1) spdo acts downstream of numb, (2) spdo null alleles transform sheaths into neurons but have minimal effects elsewhere in the SOP lineage, and (3) spdo may act likewise in adult bristle development [6, 1128, 3728]. Spdo is a homolog of tropomodulin – a protein associated with microﬁlaments. Two conjectures have been offered as to how Spdo might aid signal transduction [1128]: (1) Spdo could be part of a multiprotein complex (including N) at the cortex, or (2) Spdo could act in a cytoskeletal process that transports ligand-activated (truncated) N to the nucleus. (Both ideas ﬁnd support in Canoe, which colocalizes with N [1203, 4236] and has kinesin- and myosin-like domains that could mediate transport [2882, 3417, 3418].) Thus, STP 2 may

12

IMAGINAL DISCS

use Spdo in the IIIb portion of the SOP lineage: Decision to become sheath cell vs. neuron: State 0 (N = on): Dl N Spdo? {genes in Set 5} sheath. State 1 (N = off): Numb N. {Genes in Default Set 6} neuron. If the Gating Model is correct, then Numb would be acting less like a croupier (dealing cards that are stored by sentient cells and interpreted collectively after a full hand is dealt) and more like a gondolier (steering forgetful cells to destinations by choosing a path at each intersection).

Notch must go to the nucleus to function The issue of whether Notch acts directly in the nucleus was debated for years [1891, 2222, 3153, 4244, 4823] and has recently been settled in favor of a nuclear role [443, 446, 711, 1613, 2514], although some skeptics remain unconvinced [113, 4809] and recent data support their contention that Su(H) can act without N [218, 2255]. The history of the debate is worth recounting because (1) its grittiness shows the power of scientiﬁc dialectic and (2) its reliance on imperfect techniques shows the fallibility of scientiﬁc deduction. However, readers may safely skip this section without losing the thread of the main story. The controversy began in 1993, when N-intra (N’s intracellular domain) was found to be able to enter the nucleus and activate the Notch pathway [1271, 2542, 4161]. It was unclear whether these correlated events were causally interrelated. Based on these ﬁndings, a “Nuclear Notch Model” was formulated (Fig. 2.2). It postulated that N-intra detaches (by proteolysis) from the extracellular part of N upon binding of ligand and goes to the nucleus where it relays the signal. Why else, the model’s advocates asked, would N have nuclear localization sequences (NLSs) [132, 4161]? Skeptics stressed that N is undetectable in nuclei of wild-type ﬂies [112, 368], but defenders countered that this invisibility could be due to rapid degradation triggered by N’s PEST motif [2299, 2542]. Recent work indeed conﬁrms that (1) N-intra is needed in only small amounts to carry out its nuclear function [3805] and (2) N-intra accumulates when the proteasome is disabled [3821]. Because N has no obvious DNA-binding motif of its own, the presumption has been that N-intra functions as a “co-activator” that binds Su(H) or another transcription factor to activate downstream genes [1448, 2542].

Although now considered proven (see below), the coactivator idea is still hard to reconcile with the genetics of Su(H) and deltex: 1. The ability of excess Su(H) to transform shafts into sockets [3827] is puzzling if Su(H) needs N-intra, because shaft cells should lack nuclear N-intra (due to Numb). Conceivably, excess Su(H) compensates for the dearth of N-intra by behaving as an activator all by itself. This notion is consistent with the fact that N and Su(H) elevate transcription synergistically [750, 1131], but it is contradicted by evidence that Su(H) recruits co-repressors when N-intra is absent [449, 1330, 2129, 2941]. Perhaps, a sufﬁciently high titer of Su(H) swamps the pool of co-repressors, leaving “naked” Su(H) as the predominant species, and the solo Su(H) is then free to turn on certain target genes. Other target genes, however, are turned off, so its effects must be context dependent [1329, 1330]. 2. The gene deltex (dx) was named for the resemblance of its LOF phenotype to that of Delta – namely, wing veins that widen into deltas at their tips [2561, 4778]. It is one of only four genes recovered in an extensive screen for modiﬁers of N GOF lethality [4778, 4780]. When Deltex binds N’s ankyrin repeats, it apparently displaces Su(H), which goes to the nucleus [1448, 2746]. Such displacement implies that excess Deltex should (1) sequester N in N-Deltex heterodimers, (2) prevent N-intra from joining Su(H) as a co-activator, and (3) result in a N LOF phenotype [132]. Instead, overexpression of deltex yields a N GOF phenotype [2746]. Strangely (like Brd), deltex LOF mutations rarely affect bristle cell fates, so either deltex plays no role in the SOP clone or its function is redundantly shared by another gene [1570, 4778]. N’s ability to bind Su(H) [1269, 2747, 4244] suggested that inactive N might keep Su(H) out of the nucleus by restraining it at the membrane [1269, 1448, 4244]. Indeed, Su(H) colocalizes at the membrane with inactive N in cultured cells [1269, 1307], and movement of Su(H) to the nucleus depends on N activation both in cultured cells [1269, 1307] and in vivo in socket cells [1448]. However, the restraint idea made no sense genetically. If it were true, then deleting N should liberate Su(H) to go to the nucleus (just like activating N) [711, 1131, 1613, 2299, 2746]. In fact, N null mutations disable the pathway (App. 3). Until 1997, the Nuclear Notch Model (sans the restraint amendment) fended off criticism fairly well. Active N constructs were shown to go to the nucleus, bind Su(H), and increase transcription of Su(H)-responsive

CHAPTER TWO. THE BRISTLE

13

genes [750, 1914, 2043, 2611]. Still, it was unclear how the facts ﬁt into a causal chain in situ [4580]. In 1997, several ﬁndings were reported that seemed to contradict the model, although each challenge had a technical loophole: 1. Activated N constructs that are anchored to the membrane (in ﬂy or mammal cells) are as effective at stimulating transcription of Su(H)-dependent reporter genes as constructs that demonstrably go to the nucleus [132, 1131]. (However, proteolytic conversion of anchored N to free N-intra could not be ruled out, and minute amounts of nuclear N-intra might be enough for reporter activation.) 2. Disabling both NLSs of hNotch1 (the human Notch homolog) – by replacing 2 a.a. and deleting 49 a.a., respectively – reduces nuclear localization but does not signiﬁcantly decrease transactivation of Su(H)responsive genes [132]. (However, nuclear localization was not abolished, and residual N-intra in the nucleus may sufﬁce for reporter activation.) 3. Eliminating hNotch1’s only stable binding site for the Su(H) homolog (in its RAM23 domain [2747, 4244]) does not signiﬁcantly decrease its ability to transactivate Su(H)-dependent genes [132]. (However, Su(H) can also bind hNotch1’s ankyrin repeats, so the reduction in binding may not be enough to stop coactivation.) Doubts raised by these results made an alternative “Catalysis Model” look more attractive (Fig. 2.2) [1613]. In that model, transient contact with N supposedly changes Su(H) to an activated state, and “Su(H)*” then enhances transcription of downstream genes with no further need for N [112, 1448, 1651, 2299, 4244]. Because N itself has no known enzymatic activity [1613], N might use a binding partner as the catalyst [1131]. Conceivably, N alters Su(H)’s shape to unveil a domain without covalent modiﬁcation [132, 1131]. This “hit-and-run” scenario agreed with (1) the instability of N-Su(H) binding and (2) the fact that N and Su(H) do not always colocalize [4244]. In 1998, the tide turned for the last time. Incisive studies provided direct evidence for the Nuclear Notch Model. They showed ligand-dependent proteolysis and transit of N-intra to the nucleus, as well as transcriptional stimulation by amounts of nuclear N below histological detection [443, 711, 2514, 4582]: 1. Upon binding of ligand, mNotch1 (the mouse Notch homolog) is cleaved on the cytoplasmic side of its transmembrane domain – between residues that in

2.

3.

4.

5.

the ﬂy correspond to Met-1762 and Val-1763 [3805]. Mutating the Val to Leu or Lys virtually blocks cleavage, but even the residual amount of cleaved product is able to activate transcription of a reporter gene. The quantity of nuclear N needed for such activation is so small that it is barely detectable immunologically, even when sensitivity is enhanced with multiple Myc tags on N-intra. This result explains why nuclear N had hitherto been “invisible” in vivo, and it validates the idea that trace amounts of cell-surface receptors can directly inﬂuence nuclear events [2030, 2031]. Inserting a Gal4 DNA-binding domain into the intracellular tail of full-length N produces a receptor that can activate UAS-regulated reporter genes in a ligand-dependent manner in situ [2454, 4155], whereas similar insertions into the extracellular domain do not. Hence, cleavage and nuclear transit of N-intra are natural consequences of ligand binding. Inserting domains for transcriptional activation (Gal4 or VP16) or repression (WRPW or an Ala-rich piece of Engrailed) into N’s tail, respectively, rescues or fails to rescue N null embryos – implying that N-intra directly (in the nucleus) regulates targets of the Notch pathway in situ [4155]. A cleaved, Su(H)-bound form of N that is soluble and phosphorylated was identiﬁed in ﬂy embryos in addition to full-length N [2211], and the ratio of processed to full-length N was shown to depend on (1) the presence of N’s binding domain for Dl and (2) the amount of Dl both below and above the wild-type level. The cytoplasmic domain of N was found to have an intrinsic ability (85% that of Gal4) to activate transcription at a heterologous promoter in yeast and in vivo [2211]. To conﬁrm that N is a co-activator for Su(H), a foreign activation domain was substituted for N. When VP16 was fused to Su(H), the Su(H)-VP16 product activated transcription of m8 (a natural target of Notch signaling) in N null embryos [2211].

The inescapable conclusion is that a piece of the N molecule stimulates transcription directly in the nucleus after ligand-dependent cleavage of full-length N. The implied stoichiometry of “1 N molecule: 1 unit of activation” is consistent with the dose sensitivity of the Notch pathway [118, 1797]. This situation contrasts with other transduction pathways, where the quantity of second messenger is catalytically ampliﬁed relative to the input signal [3297].

14

Cleavage of precursor (in Golgi)

binds Scale: 100 a.a.

1 LNG repeats (3x)

Ligand-induced cleavages

2 3 N-intra

C1693 C1696

Key:

NLS (2x)

EGF-like repeats (36x)

leader 11 12

24

NH2

PEST

ankyrin repeats (6x)

opa

Notch

29

COOH

*

Delta Fringe

*

RAM23

Dsh

Serrate NH2

2703 a.a.

COOH

*

*

Notch

Su(H) Deltex N

C

Numb PTB

556 a.a.

CHAPTER TWO. THE BRISTLE

How does Numb ﬁt into this scheme? An obvious possibility (stated above) is that Numb stops Notch signaling by tethering N-intra to the cortex, and Numb does bind N-intra in vitro [1307, 1651]. If this were the whole story, then freeing Numb from the cortex should let N-intra function in the nucleus (hence shifting State 1 cells to State 0), but as the following experiments demonstrate, this prediction is not fulﬁlled. 1. Deleting Numb’s “mooring cleat” (at its N terminus) releases it into the cytoplasm but does not affect its ability to switch cell fates (by inhibiting N) when overexpressed [2269]. 2. The mouse protein “Numblike” resembles Numb except that it is cytoplasmic [4876], and expressing a Numblike mouse transgene in numb null ﬂies shuts off the Notch pathway, despite Numblike’s freedom from the cortex. 3. Numb can suppress N in cultured cells even when both overexpressed proteins are cytoplasmic [1307]. Indeed, Numb appears to block N function by sequestering N-intra in cytoplasmic heterodimers, thus keeping it out of the nucleus. While Numb’s normal function could still be to tether N, it must be able to stiﬂe N by other means. N’s Numb-binding domain contains a nuclear localiza-

15

tion sequence (Fig. 2.3), so Numb could block nuclear entry of N-intra by masking this site. Alternatively, because N’s Numb- and Su(H)-binding sites overlap [1307, 1651, 2747, 4244], Numb-N heterodimers may be unable to dock with Su(H) even if some of them enter the nucleus. (Numb does not bind Su(H) directly [4542].) A third possibility is that Numb obstructs access of the liganddependent protease [4523]. One clue to this mystery is that microtubules must be intact for Numb to muzzle N [4523]. In summary, the physical basis for the “Numb N” and “Dl N Su(H)” interactions appears to be as follows. Numb masks one or more of N’s functional domains. In Numb’s absence, binding of Dl to N releases N-intra from the membrane by proteolytic cleavage, allowing N-intra to enter the nucleus, where it functions as a co-activator with Su(H) or other transcription factors to modulate the expression of downstream genes.

E(spl)-C genes are Su(H) targets but play no role in the SOP lineage In vitro, Su(H) protein binds the nucleotide heptamer GTGG/A GAA (where “G /A ” means a G or A at this position) [492, 1131, 2453, 4408]. This heptamer resides in the promoters of at least 4 of the 13 genes in the Enhancer of split Complex (E(spl)-C) at 96F11–14 – namely, mγ , m4, m5,

FIGURE 2.3. Notch, its domains, and some of its binding partners.

Part of a cell is shown (gray = cytoplasm; microvilli at top) with two Notch (middle) and one Numb molecule (below) drawn as bars (cf. scale at upper left). Domains (variously shaded or hatched) are mapped (cf. App. 1). Other proteins (black rectangles) are not to scale. Partners are linked by hooks (cf. key), and binding sites are delimited by dashed ovals or half ovals. Despite their regularity, the various EGF-like repeats play different roles [459, 4601, 4823], and no two are identical [2182, 2542]. Delta and Serrate both bind EGF-like repeats 11 and 12 [459, 985, 3544], although they also rely on repeats 24–26 (not shown) [2419], and binding alone may not sufﬁce for activation [1943, 1944]. Fringe (cf. Ch. 6) binds repeats 24–29 [990] and at the LNG domain [2096]. Scabrous requires repeats 19–26 for its association with N (not shown) [3456], although any binding must be indirect [4601]. The famous allele split (whence the link to the E(spl)-C was deduced [3028]) is due to a missense mutation (Ile-to-Thr) in repeat 14 ts1 has a missense mutation (Gly-to-Asp) in repeat 32 (a.a. 1272) [4779]. [1747, 2182], and the widely used heat-sensitive allele N Notch is cleaved at three sites (zigzag lines) [491, 2995, 4809]. Cleavage at site 1 occurs during maturation in the trans-Golgi along the secretory pathway [368, 2299], and the fragments stay together, possibly by disulphide bridges at C1693 and C1696 [3153] (but see [446, 578]). Cleavage at site 2 occurs after ligand binding [491, 2995], releasing an ectodomain that is “swallowed” by the ligand-bearing cell [2263, 3271] (cf. the reverse with Boss [601, 2318, 4211]). The smaller size of the extracellular vestige (now <300 a.a.) stimulates a Presenilin-dependent protease to make the next (ﬁnal?) cut [3533, 4156, 4582]. Cleavage at site 3 occurs in the transmembrane domain (a.a. 1746–1765) between M1762 and V1763 [2995, 3805], releasing an intracellular fragment (N-intra) that goes to the nucleus and turns on target genes as a co-activator with Su(H) [4155] (see text for evidence). For review, see [2994]. Numb, when present, apparently tethers N-intra to the membrane (cf. Fig. 2.2). Half of Numb’s PTB domain binds Notch at RAM23, where Su(H) also binds [1307, 1651, 2747, 4244], and at a less crucial C-terminal site [1139, 1307, 1651]. Deltex may displace Su(H) from Notch when Delta is absent [112, 1448, 2746, 3022], although binding sites for Deltex and Su(H) do not overlap [2747, 4244]. (Deltex switches cell fates when overexpressed but has no LOF effect on bristles [2746].) Notch is thought to dimerize via cysteines (C1693 and C1696?) [2209, 2542] but may also do so via its ankyrin repeats [2747] or opa motif (links not shown) [3353]. Binding of Dishevelled (Dsh) to the C-terminal tail [151] short circuits the Notch and Wingless signaling pathways [356, 3690]. Notch’s other binding partners include Wingless itself [4601] (see App. 3 for other modulators). Sites of phosphorylation [368, 2209, 2211, 4582] and glycosylation [2070, 4611] are not shown. Fringe is presumed to glycosylate Notch at O-linked fucose sites in EGF-like repeats 3, 20, 24, 26, and 31 [2904]. Adapted from [368, 446, 2210, 2542, 4611]. See [4602, 4603] for apparently heretical modes of Notch signaling. See also App. 7.

16

IMAGINAL DISCS

a

E(spl)-C

-30 kb

m

-20

m

m

* * *

b70 kb

60

y

-10

m

m1 m2

50

40

M3 (7-69) M5 (14-74) M7 (9-69) M8 (6-66) M β (9-71) M γ (11-73) M δ (11-73)

30

sc

*

m3 m4

*

20

* **

10

0

l’sc

-10

helix 1

-20

-30 kb

ase T1

T2

* *

basic

20 kb

10

m5 m6 m7 m8 gro

AS-C

ac

c

0

*

loop

helix 2

KTYQYRKVMKPLLERKRRARINKCLDDLKDLMVECLQQEGEHVTRLEKADILELTVDHMRKLK KTQHYLKVKKPLLERQRRARMNKCLDTLKTLVAEFQGDDA--ILRMDKAEMLEAALVFMRKQV KTYQYRKVMKPLLERKRRARINKCLDELKDLMAECVAQTGD--AKFEKADILEVTVQHLRKLK KTQIYQKVKKPMLERQRRARMNKCLDNLKTLVAELRGDDG--ILRMDKAEMLESAVIFMRQQK KTYQYRKVMKPMLERKRRARINKCLDELKDIMVECLTQEGEHITRLEKADILELTVEHMKKLR KTYQYRKVMKPMLERKRRARINKCLDELKDLMVATLESEGEHVTRLEKADILELTVTHLQKMK KTQHYRKVTKPLLERKRRARMNLYLDELKDLIVDTMDAQGEQVSKLEKADILELTVNYLKAQQ

conserved:

Hairy (27-89)

PLKSDRRSNKPIMEKRRRARINNCLNELKTLILDATKKDPARHSKLEKADILEKTVKHLQELQ 6

13

d

e WRPW

WRPW

Gro

CACnAG

Su(H)

CAnnTG

+ ? + + f TGTGAGAAACTTACTTTCAGCTCGGT TCCCACGCCACGAGCCACAAGGATTGTCCTCCGTCCTACGAAGTTGCAGCTGT C g

S

-700 bp

-600

NE

-500

S?SNN E

-400

-300

-200

m8 -100

+1 bp

CHAPTER TWO. THE BRISTLE

and m8 (Fig. 2.4) [171, 2453]. In vivo, reporter genes attached to these promoters are activated by Su(H), and the activation requires the heptamers [171, 2453]. The implication is that E(spl)-C genes are targets of Su(H) during Notch signaling (i.e., Dl N Su(H) “Transcribe E(spl)-C genes”). Indeed, these same promoters are responsive to activated N in vivo [171]. Some connection between E(spl) (a.k.a. m8) and N had long been presumed because a mutant allele of E(spl) enhances the phenotype of split – an allele of N (hence the gene’s name) [3028, 3896]. Seven genes in the E(spl)-C contain a “basic helixloop-helix” (bHLH) motif (Fig. 2.4) [2246, 2274]. For bHLH proteins in general, the “basic” domain mediates DNA binding and the helices mediate dimerization [1152, 2634]. Some bHLH proteins form homodimers, others form heterodimers with preferred partners, and still others can do both [68, 245, 2039, 3028, 3029, 3375]. A separate cluster of four bHLH genes – the achaetescute Complex (AS-C) at 1B1–7 – also governs bristle development (cf. Ch. 3), and certain proteins from the two complexes can heterodimerize [1484, 3171]. Whereas AS-C bHLH proteins have canonical bHLH features [348, 1152,

17

3375], E(spl)-C bHLH proteins are unorthodox in several respects (Fig. 2.4) [135, 3171, 4318]:

1. Their dimers typically bind an “N box” consensus nucleotide sequence CACnAG (where “n” denotes a variable nucleotide), instead of the CAnnTG “E box” that characterizes other bHLH proteins. Nevertheless, E(spl)-C dimers can compete with AS-C dimers at E boxes [2054], and ﬂanking bases are also relevant. 2. Their basic domain contains a conserved proline, although substituting a different amino acid does not alter their afﬁnity for N boxes [3171]. 3. Their carboxy terminus invariably ends with the sequence WRPW (tryptophan, arginine, proline, tryptophan). These features deﬁne a distinct bHLH subfamily that also includes the genes hairy and deadpan [331, 1017, 2274, 3697, 3759]. The WRPW tetrapeptide binds the ubiquitous protein Groucho [2064, 3278] – a “co-repressor” (i.e., an inhibitory factor that “piggybacks” on DNA-binding proteins). Groucho reduces transcription by recruiting a histone deacetylase [737]. Interestingly, the groucho gene resides in the E(spl)-C. By bringing Groucho to N boxes,

FIGURE 2.4. Clusters of bHLH genes and the roles of their proteins as transcriptional regulators.

a, b. Enhancer of split (E(spl)-C; a) and achaete-scute (AS-C; b) complexes. Genes are depicted as arrows, indicating the direction of transcription. E(spl)-C spans ∼50 kb on the 3rd chromosome; AS-C spans ∼100 kb near the tip of the X. Asterisks mark genes whose products have a bHLH motif. Oddly, all these genes are devoid of introns. Within each complex, the bHLH genes exhibit partial functional redundancy [2548, 4674]. AS-C bHLH proteins are transcriptional activators, whereas E(spl)-C bHLH proteins are repressors. Half circles mark E(spl)-C genes that belong to a separate (“Bearded”) gene family [2383]. c. Amino acids in bHLH motifs of E(spl)-C proteins and Hairy (AS-C proteins not shown; for code, see App. 1). Dashes are inserted to aid alignment [2274]. Numbers are residues counted from the N-terminus. E(spl)-C bHLH proteins are ∼200 a.a. long, with ∼60 a.a. in their bHLH domain. The “conserved” row pertains to E(spl)-C proteins: ﬁlled circles (invariance); unﬁlled circles (>50% but <100% identity); +s or −s (charged residues). The many +s deﬁne the “basic” domain, which, in this subfamily, has proline at position 6 and arginine at position 13 [3179]. d, e. bHLH proteins dimerize via helical domains (striped), adopt a scissors shape, and bind DNA via basic domains (black). Within each subunit, the upper and lower helices touch (unlike in this cartoon). d. E(spl)-C bHLH dimers bind an “N box” consensus sequence “CACnAG” (“n” is a dimer-speciﬁc nucleotide) [3171, 3759], which Hairy can also bind, although it prefers CACGCG [3179, 4451]. Their C-terminal “tails” (top) end in “WRPW” (as does Hairy’s [4524]) [2274, 3697], which recruits Groucho (Gro, black rectangle). Gro is a co-repressor (X’d arrow) [2064, 3278] that in turn binds a chromatin-modifying histone deacetylase [737], although E(spl)-C proteins also have Gro-independent effects [3029]. The gro gene resides in the E(spl)-C (a). e. Most other bHLH dimers (including AS-C) bind an “E box” hexamer “CAnnTG” [348, 3375], do not end in WRPW, and activate transcription (cf. the bHLH-PAS subgroup [4022, 4612]). f, g. Promoter region of the m8 (a.k.a. E(spl)) gene. f. Details. Binding sites (S = Su(H); E = E box; N = N box) are boxed (f) or shown as bars (g). “?” denotes a hexamer (function unknown) found between invertedly repeated “S” sites [171, 3075], whose interrepeat distance is constant among promoters. +s and −s signify stimulation vs. inhibition of transcription. g. “Wide-ﬁeld” view of the m8 cis-enhancer region. Dashed rectangle marks the section shown in f – viz., bases −133 to −211 b.p. from the transcription start site (“+1 b.p.”; right-angle arrow). Negative feedback of m8 onto its own N boxes [871] may help to stabilize output and minimize noise [262, 1383, 3347]. The crowding of binding sites suggests steric competition or “quenching” [1984] that mediates “either/or” (vs. “both/and”) logic [2350]. Additional N boxes and “S” sites (of varying afﬁnity) map in the 700 b.p. span (g) [2317, 3075], although we do not know whether they are all needed [449]. The base at −185 (f) may be T [171] or C [2246]. Cis-regulatory regions for all E(spl)-C genes (except gro) have been analyzed similarly [871, 3075]. Maps of loci were compiled from Fig. 3 of [3806] amended as per [2383,4767] (a) and from [636, 1538] (b). Sequences in c are from [2274]. Panels d and e are based on [1152, 2634], and panels f and g are adapted from [171, 2246, 2317, 3171]. See [2054] for an exegesis of E vs. N boxes and an exploration of target gene preferences for AS-C vs. E(spl)-C proteins. See also App. 7.

18

IMAGINAL DISCS

E(spl)-C bHLH proteins should repress the transcription of any genes that have N boxes in their promoters. One such gene is m8, which hence should repress itself, although the proximity of its N boxes to Su(H)binding sites (Fig. 2.4g) suggests that autorepression is prevented by competitive binding during Notch signaling [171]. If so, it seems odd that SOPs express a lacZ reporter linked to an m8 promoter (ectopically) when its Su(H)-binding sites are deleted [2453], but the promoter could still respond to AS-C bHLH proteins that bind its E boxes [171, 3171]. Phrasing the logic at m8 promoter as an imperative, the rule would be roughly as follows: “Turn on IF you are occupied by Su(H) or AS-C proteins, but NOT IF there is too much of your own gene product around, in which case you should remain off.” Another way of thinking about the output of m8 is as an equation whose inputs – Su(H), AS-C, and M8 – have weighting factor coefﬁcients. This glance at m8 serves to show (in microcosm) the intricacy of the control system. We do not yet know which (if any) E(spl)-C genes are expressed in the SOP progeny [991]. Genetic analysis has been difﬁcult [2561, 4885] due mainly to the functional redundancy of the bHLH genes [991, 1017, 2548, 3806]. Indeed, deletion of m8 (whose neomorphic E(spl)D allele had long implied a key role for this gene [4318]) has no detectable phenotypic effect [1018, 3806]. Overexpression of wild-type E(spl)-C bHLH genes can cause various changes in cell fates [2317, 4256], but these abnormalities may be misleading because some of them persist when conserved domains (bHLH and WRPW) are deleted [68, 1475]. The biologically relevant question is: Do E(spl)-C bHLH proteins normally assign fates in the bristle organ? Given the following data, the answer seems to be “No”:

sion of the chimeric transgene (via Gal4-UAS control) increases bristle density, but fails to switch cell fates in the bristles themselves [2063], implying that m7’s downstream targets are irrelevant to the SOP lineage. 4. Overexpressing the bHLH genes mδ, mγ , or m7 suppresses SOP initiation at many sites, but the remaining bristles have a normal set of constituent cell types [3027]. When either of the non-bHLH genes mα or m4 is overexpressed, it increases bristle density (by inhibiting bHLH E(spl)-C genes) but fails to alter the SOP lineage [92].

1. Deletion of all 7 E(spl)-C bHLH genes in somatic cell clones increases the density of bristles (for reasons discussed in Ch. 3), but has virtually no impact on bristle cell identities [991, 3027]. 2. Deletion of various E(spl)-C bHLH genes rescues the bristle-loss phenotype of null mutations in Hairless (an antagonist of the Notch pathway, see below), but does not alter the shaft-to-socket transformation also caused by these mutations [197]. 3. Replacing the WRPW (Groucho-binding) site of M7 with a transcriptional activator produces a chimeric M7 protein that should bind N boxes (since it still has its bHLH domain) but activate (not inhibit) its gene targets (due to the C-terminal substitution). Expres-

The transcription factor Tramtrack implements some cell identities

Only one bHLH gene remains a viable candidate for the setting of cell fates within the bristle organ during normal development – the AS-C gene asense. Asense is found in SOPs, and its expression persists into IIa and IIb, although no asense mRNA has been detected beyond this point [438, 1079, 2039]. Given that all SOPs make Asense, it is surprising that only one group of bristles is affected when the gene is deleted: a row of stout bristles along the wing margin [2039]. In asense null ﬂies, this row manifests empty sockets, missing bristles, stunted shafts, and twinned shafts. Twinning could indicate a socket-to-shaft or IIb-to-IIa conversion, although neither is certain because the SOP lineage here (in wildtype ﬂies) can violate the rule of ﬁxed fates (e.g., socket and shaft cells do not need to be sisters) [1741]. When the asense deletion also removes certain AS-C enhancers, a sheath-to-neuron switch is seen in the giant sensillum of the radius [1079]. The importance of asense inferred from this LOF data is not upheld by the GOF data. Overexpression of Asense (via a heat-shock promoter) induces extra bristles, but it fails to alter any cell fates within the bristles themselves [438, 1079].

The gene tramtrack (ttk) was recovered in screens for proteins that bind enhancers at the ftz pair-rule locus [502, 1731]. Its transcripts are spliced to produce two protein isoforms – “p69” (69 kD) and “p88” (88 kD) [3537]. Both isoforms typically function as transcriptional repressors [503, 1483, 4596, 4773], but p69, unlike p88, becomes an activator in late eye development [2391]. Both proteins have an N-terminal “BTB” (protein interaction) domain and two DNA-binding zinc ﬁngers, but the ﬁngers of p69 and p88 differ in amino acid sequence (and hence target DNA sequence) because they come from different exons (Fig. 2.5) [3537].

CHAPTER TWO. THE BRISTLE

a3 kb

19

tramtrack gene 4

5

6

7

8

9

10

ZnFs

A

B

A

B

11

12 kb

ZnFs

C

p69 D

p88

b

p69 (ZnFs): YRCKVCSR-----VYTHISNFCRHYVTSHKRNV-KVYPCPFCFKEFTRKDNMTAHVKIIHKI p88 (ZnFs): YRCTECAKENMQKTFKNKYSFQRHAFLYHEGKHRKVFPCPVCSKEFSRPDKMKNHLKMTHEN conserved:

Asn

p69 protein

' ' 5-ATCCT-3

d

H

C C

N

H

O

H H

C N

Adenine

N

N C

C N

C N

C

H

ine

H

ym

' ' 3-TAGGA-5

H

Th

c

YR K RNVKVYP C HS KI P C K T H Zn F I V V Zn I C C HY K F HV S R K A R C E T V F F M Y N T N T S R KD H I

H

FIGURE 2.5. Alternative splicing of tramtrack yields different DNA- binding proteins.

a. Transcripts (fragmented arrows) of the ∼9-kb gene tramtrack (ttk, 100D) encode two proteins, each of which has different zinc ﬁngers (“ZnFs”) [3537, 4773]. Kinked lines are spliced-out pieces, and lightly shaded bars are untranslated sequences. The 69-kD “p69” isoform (641 a.a.) has sections A, B, and C. The 88-kD “p88” isoform (811 a.a.) has sections A, B, and D. Shared segments (A, B) encode a “BTB” protein-interaction domain [211, 1516, 4891], and unshared ones (C, D) encode zinc ﬁngers. b. Amino acid sequences of the ﬁngers. White letters are Cys and His residues that coordinate Zn2+ ions, dashes are gaps to aid alignment, and black circles mark identical residues. c. Zinc ﬁngers (shaded) of p69 are drawn ﬂat, although each outlined half is actually an α-helix, and the other half forms β-strands. p69 binds DNA as a monomer [1186], inserting the α-helices of its ﬁngers consecutively into DNA’s major groove. Arrows indicate a.a.-base contacts. The DNA oligomer used for structural analysis was devised from p69’s enigmatic set of binding sites [1186]. p88’s ﬁngers (not shown) bind other sequences [3537, 4775] and have not been studied crystallographically. Transcriptional repression (lines radiating to X’d arrow) might be due to a C-terminal domain (BTB is N-terminal). In some unknown way, p69 (but not p88) is converted from a repressor to an activator in eye development, where the two isoforms act differently in different cell types [2391, 2529]. Both isoforms are regulated by proteolysis [1051, 2529, 4249]. d. Hydrogen bonds (shaded) between asparagine and adenine (cf. N-to-A arrows in c) vs. bonds within the A-T base pair (at right) revealed by crystallography [1187]. Such interlocks have fueled hopes for a “ﬁnger code” that could allow ﬁngers to be designed for desired DNA sequences [296, 2265]. This diagram is adapted from [4773] (a), [3537] (b), [1187, 2123] (c), and [296] (d).

20

This use of alternative splicing to send transcription factors to different gene destinations is a clever sort of grammar in cell programming [345, 2599, 3996]. The imperative “Turn off!” (or in some cases “Turn on!”) can thus act on various genes via different “Go to” addresses (the ﬁngers of p69 vs. p88) [4339]. This “Finger Shufﬂing Trick” is used by at least two other ﬂy genes: (1) the “pufﬁng cascade” gene Broad-Complex, whose BTB domain is spliced to one of four pairs of zinc ﬁngers [247, 248, 1044, 2974] in a spatially regulated manner [458], and (2) a (nonBTB) chorion transcription factor [1519, 1918]. Alternative splicing is also used for other types of “Go to” commands, which (1) add or delete NLSs that send proteins to the nucleus vs. cytoplasm [957, 2032], (2) insert a peptide that sends proteins to intercellular junctions [4574], and (3) may even direct the wiring of sensory axons to speciﬁc CNS neurons [4763]. Deleting ttk function in somatic clones causes the same 4-neuron trait seen in N LOF ﬂies, but overexpressing ttk (via a heat-shock promoter) only rarely yields the 4-socket trait of N GOF [1650]. Instead, the major GOF phenotype is “2 shafts, 2 sockets” [1650, 3502] – indicative of a conversion of IIb into IIa. (Strangely, the GOF phenotypes for p69 and p88 are similar despite their different binding-site preferences.) Hence, ttk may implement the IIa/IIb decision. Its null phenotype suggests that ttk also mediates the sheath/neuron decision because two switches must occur for an SOP to make 4 neurons (IIa-to-IIb and sheath-to-neuron). Consistent with this reasoning, p69 is detectable in IIa (not IIb) and in sheath cells (not neurons) [3502]. The rarity of 4-socket or 4-shaft defects implies that ttk plays no role in the shaft/socket decision – a conclusion bolstered by Ttk’s presence in both shaft and socket cells [3502]. Thus, as a rule, ttk is expressed in non-neural cells (or their ancestors) in the SOP lineage [3502]. Similar rules govern ttk function in the CNS and eye, so ttk may be a generic repressor of neural fates [1483, 4773]. Ttk may be kept off in IIb by proteolysis [3550] because (1) in the eye Ttk is targeted for degradation when it binds Seven in absentia (Sina) and Phyllopod (Phyl) [4249], and (2) LOF mutations in sina and phyl cause doubled bristles [671, 717] that suggest a IIb-to-IIa conversion [3550]. If ttk toggles cell fates in the IIIb sublineage (sheath vs. neuron) in the same way that Su(H) behaves in the IIa sublineage (socket vs. shaft), then is Ttk the counterpart of Su(H)? Probably not, because Ttk has not been shown to bind N (whereby it could form a factor-cofactor com-

IMAGINAL DISCS

plex). Other candidate transducers for the IIIb sublineage have been recovered [6].

Hairless titrates Su(H) Given the involvement of both Su(H) and Ttk in the IIa/IIb decision, but their separate roles in the IIa and IIb sublineages, there may be at least three transduction modes. Hairless (H) [200, 2659] also ﬁts into this scheme. Until 2000, its effects were thought to be conﬁned to the socket/shaft decision, but a detailed LOF-GOF analysis proved its participation in all three decisions [3027]. Removing H forces cells into State 0, while overexpression forces them into State 1 (App. 3). STP 1 : Decision to become IIa vs. IIb: State 0 (N = on): Dl N Su(H) {ttk, etc., but not E(spl)-C} IIa. State 1 (N = off): Numb N. H and {genes in Default Set 2} IIb. STP 2 : Decision to become socket vs. shaft cell: State 0 (N = on): Dl N Su(H) {genes? [not ttk or E(spl)-C]} socket. State 1 (N = off): Numb N. H and {genes in Default Set 4} shaft. STP 3 : Decision to become sheath cell vs. neuron: State 0 (N = on): Dl N Spdo? {ttk, etc., but not E(spl)-C} sheath. State 1 (N = off): Numb N. H and {genes in Default Set 6} neuron. Moderate H LOF alleles delete shafts (hence its name), either by blocking SOP initiation (“bristle loss”) or by converting shaft cells into socket cells (“double-socket” phenotype) [198]. Macrochaetes (large bristles) are affected more than microchaetes (small bristles), and certain sites are consistently affected more than others (Fig. 2.6a) [997, 2690]. Why should similar bristles differ in their need for the same gene product? This “Nonequivalence Riddle” has nagged theorists not only for H LOF [3053, 3054] but also for other “missing bristle” mutants [3405] – especially scute LOF whose subpatterns are allele speciﬁc [650, 767, 1108, 1328, 4100] – and for “extra bristle” mutants [733, 3028]. Topological models (wherein particular genes control particular bristles in 1:1 correspondence) were discounted because (1) phenotypes are so easily modiﬁable by temperature or overcrowding [767--769, 1997, 1998, 3405] and (2) double-mutant combinations can remove bristles outside the areas affected by either mutant alone

CHAPTER TWO. THE BRISTLE

[3405, 4187]. For the AS-C, the spatial heterogeneity is due to each bristle’s reliance on the sum of (functionally redundant) Achaete and Scute, whose amounts are set by position-speciﬁc enhancers (cf. Ch. 3). Hairless may likewise have a partially redundant “partner” in some body regions because leg bristles remain virtually normal even in H null mutants [198]. Although the reason for H ’s spatial nonuniformity is unclear, the basis for its quantitative effects is understood, thanks to a peculiar feature of this locus – namely, its “haplo-insufﬁciency.” Only 21 genes in the ﬂy genome (aside from Minutes) manifest haplo-insufﬁciency (i.e., display a mutant phenotype when present in haploid dose as a deﬁciency/+) [118]. Presumably, the products of such genes are “limiting” for the rate of particular reactions [117] or for the stoichiometry of particular partnerships [2582, 3458]. In the case of Hairless, the haplo-insufﬁciency is attributable to a physical coupling between the H and Su(H) proteins. As the name indicates, Suppressor of Hairless LOF mutations suppress H LOF phenotypes. The suppression can be mimicked by lowering the dose of Su(H). Thus, heterozygous H LOF/+ ﬂies that are also heterozygous for a Su(H) deﬁciency look nearly wild-type [117]. Conversely, increasing the dose of Su(H) proportionally enhances the phenotype of H LOF/+ ﬂies, as expected for a stoichiometric interaction (Fig. 2.6c). This titratable antagonism is illustrated most dramatically by the creation of a H LOF phenotype in a wild-type (H +/H + ) background by simply raising the dose of Su(H) to 8 (from its normal 2) [3827]. The identity of the Su(H) “substitute” that H opposes in STP 3 (sheath cell vs. neuron choice) remains to be determined [3027]. H is detectable in all SOP descendant cells [3027], but subtle differences in its levels may still exist in State 0 vs. State 1 cells. Alternatively, H may be modiﬁed posttranscriptionally by auxiliary factors in one or the other cell state. Hairless is a weird protein. It is extremely basic (overall pI = 9.5; pI in parts = 11, based on a.a. composition; Fig. 2.6d), although it has no known DNA-binding motif. It is 40% alanine, serine, or proline [200, 2657]. Its function in ﬂies seems to be to stiﬂe Notch signaling during bristle development (SOP lineage and SOP initiation) [3027, 3824], but less so during the other processes where Notch is deployed [113, 444]. Su(H) and H bind each other in yeast two-hybrid assays [171, 1332, 2657], and their interfaces have been identiﬁed [492, 2695]. Because the H- and DNA-binding domains within Su(H) overlap, H might block Su(H) by masking its DNA-binding site [492]. How-

21

ever, truncated H transgenes that contain the Su(H)binding domain do not fully rescue H phenotypes in vivo [2657], so H must do more than just dock with Su(H) [1329, 2658]. Indeed, it has recently been shown that the Su(H)-H complex binds DNA and that H recruits the co-repressor “dCtBP” (see Fig. 3.12e and [2129] sequel). Why should any antagonist for Su(H) be needed – titratable or otherwise? Shouldn’t shaft cells have a silent Notch pathway after Numb has gagged N? Perhaps, but the pathway may not be silent enough. Residual N-intra or Su(H) might linger from the mitosis that created IIa. H’s role could be to “mop up” excess Su(H) to ensure that prospective shaft cells do not hear any N signal whatsoever and hence become shafts with 100% ﬁdelity. H could set a constitutive threshold above which Su(H) must rise in order for a cell to become a socket [197, 3826]. A similar damping step has been proposed for gating N-Su(H) heterodimer entry into the nucleus [2211]. In principle, thresholds are an easy way to convert messy (analog) signals into discrete (digital) states [3573, 4513] – as Charles Plunkett, a pioneer of bristle research, argued in 1926: The only way in which I can conceive of an ‘‘all-or-nothing’’ reaction, such as the production of a bristle, being determined by the concentration of a continuously varying substance, is that the reaction occurs if, and only if, that concentration equals or exceeds a certain ‘‘threshold’’ value. [3405]

Thresholds also help to explain “ﬂuctuating asymmetry” [2693, 2901, 4460], where the left and right sides of single individuals show different phenotypes. Such asymmetries are rare in wild-type ﬂies [2866, 4746, 4747] but common in mutants such as H LOF [1961, 3405, 3571, 3572, 4514, 4674]. For H LOF , the reason may be that the amount of H sinks to the level of Su(H) in some prospective shaft cells. In the close contests that ensue, the outcome will depend on random local ﬂuctuations (a.k.a. “noise” [1383, 2259, 2518, 4513]) in both values (cf. error bars in Fig. 2.6e). Thus, a prospective shaft cell on the ﬂy’s left side might sense more H than Su(H) and make a shaft (resulting in a normal bristle), while its counterpart on the right senses more Su(H) than H and makes a socket (yielding a “double-socket” phenotype). Because the same is true for H-deﬁciency heterozygotes, such asymmetries cannot be dismissed as being due to “leakiness” of alleles [767, 1998, 3067]. Rather, they are symptomatic of a breakdown in the “robust buffering” of the fate-assignment mechanism itself [1440, 3845, 4513].

22

IMAGINAL DISCS

Macrochaetes

a b

Key

SC PA NP PS HU VT PV OC OR DC SA P A P A P A P A P A D V P A P M A

4

5

100

H /+

3

% Present

s

1

0 100

w

H /H

2

w 0 100

w

H /H

s 0

d

c

basic

NH2

40

1059 a.a.

NLS (3) basic basic

COOH

*

**

Def./+

594 a.a.

Su(H)

+/+

COOH

* opa (2) * NLS (2)

e

20

100 a.a.

opa

Threshold Model

Dup./+

Su(H)

10

1

1 H

0

0 1

2 3 Doses of Su(H) +

4

Socket

Activity

Dup./Dup.

Activity

Number of macrochaetes in H s/+ flies

DNA-binding NH2

30

H

PRD

H Su(H)

0

Shaft

CHAPTER TWO. THE BRISTLE

Interestingly, shaft cells in H hypomorphs can display any morphology on the spectrum from shaft to socket (Fig. 2.6b) [198]. Likewise, overexpressing H during shaft morphogenesis (via a hs-promoter) can drastically deform the shaft: Unexpectedly, heat shocks after bristle determination affected morphogenesis of shaft and socket cells of macro- as well as microchaetae. At an early stage, very frequently shaft trifurcations were observed at the expense of the socket, the central shaft always being the longest. Later shocks caused a thickened shaft base with one to several spines. . . . If shocks were applied even later, shafts were completely absent, but instead, misshapen sockets were observed. This bald phenotype is inducible over a very long period until rather late stages of [the pupal period]. [2657]

Conversely, overexpressing N [3270] or Su(H) [3827] during that period partially transforms shafts into sockets, making them shorter and thicker. Intermediate shapes such as these are probably due to close contests where cells cannot select fates decisively. The intergradations also suggest that the “obvious” dissimilarity of shafts and sockets is deceptively superﬁcial. They may actually be outputs of the same dendrite-wrapping algorithm [1741, 2173]. If so, then shafts and sockets would actually differ quantitatively (e.g., in their extent of elongation). Little is known about the molecular machines that cells use

23

to modulate their shapes, and even less is known about the genetic circuitry that runs them (see below).

Several other genes help determine the 5 cell fates The aim of this chapter is to chart as straight a path as possible from Numb, which ordains cell fates from its perch at the cell membrane, to the genes that must implement those fates. Six players have been proﬁled (Numb, Delta, Notch, Su(H), Hairless, Tramtrack), a contender has been considered (Asense), and one putative set of implementing genes has been ruled out (the E(spl)-C). Below, this same route is retraced from surface to nucleus, with ﬁve more actors added, whose roles are less central: Serrate, Nak, Twins, Musashi, and Shibire. Until 1998, Dl was thought to be the only ligand for N in bristle development. This notion stemmed from the similarity of phenotypes in heat-sensitive Dl LOF and N LOF mutants. Both kinds of mutants exhibit the same “balding” (superﬁcial bristle loss) when exposed to pulses of high temperature around the time of SOP mitoses, and in both cases the 4-neuron clusters at many denuded sites indicate switching of IIa to IIb and sheath to neuron [1742, 3272]. However, when null alleles are made homozygous in somatic clones, Dl and N do not give the same phenotype: N null clones are bald, whereas Dl null clones make tufts of relatively normal bristles

FIGURE 2.6. Effects of Hairless mutations on bristle cell fates (a, b), and interactions between Hairless and Suppressor of Hairless

(c–e). a. Half head and thorax of a wild-type ﬂy (above), with all bristles omitted except macrochaetes. Beneath, macrochaetes (black circles) are seriated in posterior-to-anterior order, except that some are grouped [2560] as scutellars (SC), postalars (PA), dorsocentrals (DC), supra-alars (SA), notopleurals (NP), humerals (HU), verticals (VT), or orbitals (OR). Relative positions of intragroup members are posterior (P), anterior (A), dorsal (D), ventral (V), or middle (M). Presutural (PS), postvertical (PV), and ocellar bristles (OC; oval is an ocellus) are unpaired. The histograms show how each site is affected by Hairless genotypes. “H s ” and “H w ” are strong and weak LOF alleles of Hairless, respectively (“+” = wild-type). Black bars are frequencies (percent) of “normal” bristles (shaft and socket); shaded bars are frequencies of “double sockets,” which likely arise from a shaft-to-socket cell transformation (key at right). H function is evidently more limiting in shaft cells than in incipient SOPs because mild reductions in H levels (uppermost histogram) cause more double sockets than absent bristles [198]. The variation in sensitivity among sites makes no sense in terms of their seriation. b. Intermediate phenotypes between a normal bristle (type 1) and a “double socket” (type 5). In type 2, the shaft has abnormal ﬂuting and pigmentation, and the socket fails to form a complete circle around the base. In types 3 and 4, the shaft is a vestige that ultimately (in type 5) resembles a stunted socket. c. Number of “normal” (shaft and socket both present) macrochaetes on the head and thorax of H LOF /+ ﬂies (H 1 allele) carrying varying doses of the wild-type allele of Su(H). “Def.” and “Dup.” are a deﬁciency and duplication for the Su(H) locus. d. Domains in the H and Su(H) proteins (cf. App. 1). Dashed ovals connect regions needed for H-Su(H) binding [492]. “Basic” here means a segment of the protein where the average pI is 11 (sic!) [2657]. The DNA-binding domain of Su(H) is probably bipartite (i.e., same limits but centrally inert), given the properties of its mouse homolog [788]. e. Threshold Model of Bang et al. [197]. In this model, H and Su(H) titrate one another, and whichever protein remains undimerized determines cell fate. That is, excess Su(H) causes a cell to become a socket, and excess H causes it to become a shaft. N.B.: In other contexts, Hairless may act independently of Su(H) [725]. Data in a are from Table 2 of [198], b is a schematic of that article’s Fig. 5, c is replotted from [117], d incorporates data culled from [492, 1331, 2657, 3826], and e is based on [197].

24

(N.B.: In the original study [1797], Dl 9P39 patches made tufts, whereas Dl RevF10 patches – like N null – were bald, but a later study found no such allelic difference [4859], and both alleles may be nulls [3272].) In the absence of either N or Dl, all SOPs should have made neurons instead of bristles. How can normal bristles develop without Dl? The answer is a redundant ligand. Serrate (Ser) was known to encode a N ligand but was not suspected of acting in the SOP lineage because none of its mutant phenotypes (including null ones [4029]) involve bristle defects [2561]. However, Ser and Dl are related proteins [1252, 4302], and Ser (when artiﬁcially expressed) can substitute for Dl in embryonic neurogenesis [1634], implying an overlap in functional ability. Whereas Ser null clones look wild-type and Dlnull clones make tufts, doubly mutant Ser null Dl null clones are bald like N null and have extra neurons at denuded sites [4859]. Evidently, Dl and Ser both activate N in the SOP lineage, although Dl’s contribution is greater (and hence less dispensable) than Ser’s. (Why t.s. Dl LOF defects should be stronger than Dl null clonal defects remains unclear.) Consistent with this conclusion, shaft cells can be transformed into socket cells by overexpressing either Dl or Ser in neurons [2008] (Fig. 2.7c). A yeast two-hybrid screen [763, 1228] for Numb-binding proteins netted numb-associated kinase (nak), which encodes a putative serine/threonine kinase [765]. Nak interacts with Numb’s PTB domain, which is interesting from several standpoints: [4859].

1. Numb can localize properly without this domain [1307, 2269], so Nak probably does not escort or anchor Numb to the cortex [765]. 2. Like Nak, N binds Numb’s PTB domain [1307], so Nak might displace N from Numb, in which case it should antagonize Numb. Indeed, it does: overexpression of Nak alters cell fates in the opposite way from numb GOF [765]. 3. The PTB domain is a docking motif that binds phosphotyrosines [765, 2530], so Nak might hinder Numb indirectly by phosphorylating substrates that directly displace N. Nak’s true role will not be known until LOF alleles can be isolated, but the Numb-N mechanism must involve phosphorylation somehow because LOF mutations in twins can switch IIb to IIa just like numb LOF mutations [3904]. Twins is the regulatory subunit of serine/threonine phosphatase PP2A in D. melanogaster [4418]. Other twins LOF alleles interfere with mitosis [1535, 2761]. Such

IMAGINAL DISCS

effects are consistent with PP2A’s role in other species [830, 3352, 3904], and they suggest that Twins might be responsible for the cell-cycle dependence of Numb’s cortical localization [2267]. LOF mutations in musashi also produce numb LOF like phenotypes [3035]. This gene encodes an RNA-binding protein. Because Musashi protein is nuclear, it probably does not localize mRNA (like Staufen [478, 1321, 2526]) or regulate translation. It might control RNA processing (splicing? [193, 1100, 2186]) in a positive manner for numb, nak, twins, or Hairless, or in a negative manner for Dl, N, Su(H), or tramtrack. Indeed, Musashi has recently been found to bind and muzzle tramtrack transcripts ([3035] sequel). Ever since Numb’s cardinal role in dictating bristle cell fates was revealed in 1994 [3579], many pieces of the genetic puzzle have fallen into place. One remaining mystery [2268] is: What upstream cell-polarity cues cause Numb’s cortical asymmetry? During asymmetric mitoses, the Numb crescent always overlies one pole of the spindle [2267, 4028] . This coordination makes sense because otherwise the division plane could bisect the crescent and dispense equal amounts to both daughters [3580, 4875]. Embryonic neuroblasts (NBs) and ganglion mother cells also parcel Numb unequally. In their case, spindle orientation and Numb localization are both controlled by Inscuteable (Insc) and Miranda [4323] . Miranda is an adaptor protein that interacts with Insc and Numb [1322, 3893] . Numb’s N-terminal 227 a.a. sufﬁce for proper crescent formation in NBs [2269] and may be hooked to Insc via Miranda [3893]. However, insc LOF mutants do not show defects in bristles [3550] or larval sense organs [2327] (although IIb mitoses are misoriented [3630]), so Insc seems dispensable [2607]. Miranda is present in the IIb cell [3630] (despite an earlier report to the contrary [2680]), but its function there (if any) is unknown. Other links in the NB chain of command may also be irrelevant [2017], such as Bazooka (upstream of Insc [3803, 4709]) and “Partner of Numb” (downstream [2606]), which binds Numb and Miranda [2609]. This conclusion seems odd given the many similarities between SOPs and NBs [587, 1805] . For example, (1) IIb and IIIb divide along the apical-basal axis [1447, 3630] like NBs, which localize Insc apically [2328, 2607, 2609]), and (2) the homeodomain protein Prospero segregates basally in IIb and IIIb [1447], as in the embryonic CNS [477] , and helps decide cell fate in both systems [2680] . The current working hypothesis for Numb localization (in both SOPs and NBs) is that Numb is recruited to the cortex by the WD-repeat protein Lethal giant larvae and then driven to its ﬁnal destination by a polarized

CHAPTER TWO. THE BRISTLE

a

25

If input =

Numb and Notch

Dl Ser

Key:

Su(H)

or

and

Socket

H

Shaft x

b

0

1 and

1 1

or

1

1

1

IIa

or

0

0

1

neuron

0 or

1

1

1 1

0

0 1

Shaft

y

1

1

1

nt

1 and

u sh

0

1

Socket

and 1

1

1

nt

s

hu

x and z y

x 0 0 1 1

y 0 1 0 1

z 0 0 0 1

x y

x 0 0 1 1

y 0 1 0 1

z 0 1 1 1

or

z

intercellular

0 mutual antagonism

1 and

0

x y 0 1 1 0

1

0 or

x y 0 0 1 1

"driver" 1

1 and

1

0

mitosis

neuron

IIa

0

and 1

c

x

Socket

0 and

1 1

1

1

mitosis 1

y not

1 and

0

then output =

symbol

1

Socket

1 0

FIGURE 2.7. Circuitry that assigns shaft vs. socket cell fates in the SOP lineage. Logic symbols (key at right; cf. [2284, 4735]) denote

activation, inhibition, combinatorial criteria (“and” gate), and redundancy (“or” gate). The “not” symbol is a blend of genetic and electronic icons [1751, 2770], and the reciprocal “not” condition (“mutual antagonism,” “ﬂip-ﬂop,” or “seesaw”) means that if one partner is on (“1”), then the other must be off (“0”). These relations are summarized in the truth tables, where “1” means “true” and “0” means “false” [399, 1433]. The “driver” is the element that chieﬂy dictates the outcome, and “intercellular” connotes a ligand (arrowhead) and receptor (“V”). a. Core pathway of proteins that decide shaft vs. socket cell identities (cf. Fig. 2.2 and text for details). Abbreviations: Dl (Delta), H (Hairless), Ser (Serrate), Su(H) (Suppressor of Hairless). In plain English, this circuit means: “Activate the Notch receptor if it binds either Dl or Ser on an adjacent cell (but go no farther if Numb is present!), then let it bind Su(H) and (as long as there’s not too much Hairless around!) let the Notch-Su(H) complex turn on genes for socket identity; otherwise, turn on genes for shaft identity.” b. The circuit as it functions in daughters of the IIa cell (Fig. 2.1) in wild-type ﬂies. Numb is normally the decisive factor in setting cell fate (see text). The daughter that lacks Numb (above) has an active Notch pathway and hence forms a socket, while the daughter that inherits Numb (below) has an inactive Notch pathway and hence forms a shaft. c. The circuit as it malfunctions when neurons are forced (via neuron-speciﬁc Gal4-UAS driver “31-1”) to make excess Delta [2008]. In this situation, the prospective shaft cell makes a socket instead, presumably because hyperstimulation of Notch receptors supersaturates the limited amount of Numb. The ability of such excesses to “short circuit” the system (thick line) implies that care must be exercised in interpreting GOF phenotypes (in general) whenever proteins “ﬂood” a network.

26

IMAGINAL DISCS

cut

poxn

0 0 CT

1 1 CS

1. smell

1. stretch A CNS

A

A

N N

A

N CNS N 2. "fire"

N CNS

Ato

(Prepattern)

N

N

N

N 2. "fire"

Cut SOP: CT

and

and

SOP: CS GOF

Poxn

cut or

and

LOF

LOF

poxn

SOP: MS

GOF

1 0 MS

1. touch

3.

st

re

tc

h

ato

poxn

N

2. pivot

CNS

actomyosin motor [468,3205,3326], where it is anchored by Gαi (inhibitory subunit of GTP-binding protein) [3776]. Near Numb’s C terminus is a tripeptide – NPF (asparagine-proline-phenylalanine) – that binds “EH” (Eps15 Homology) domains [3725, 4417]. Because NPF or EH motifs are found in proteins that control clathrinmediated endocytosis [2699, 2756], Numb’s inhibition of N might entail effects on endocytosis. Indeed, Numb can inﬂuence the endocytic internalization of membrane receptors [3757]. Evidence that Notch signaling requires endocytosis comes from shibire (shi). Heat-sensitive shi LOF mu-

CNS

N 4. "fire"

tations cause the same anatomical defects as N LOF mutations and share the same sensitive periods [1057, 3423, 3425, 3863, 3891]. Among the defects are socket-to-shaft transformations [1803, 3425]. The protein encoded by shi is the ﬂy homolog of the mammalian GTPase “dynamin” [740, 4443]. In neurons, dynamin self-assembles around the necks of clathrin-coated pits, and a conformational change in these “collars” appears to pinch off vesicles from the membrane, although the mechanism may involve pushing instead of pinching [2795] . Detachment of vesicles halts in t.s. shi LOF ﬂies at high temperature [1016] , which explains why shi was recovered in a screen

CHAPTER TWO. THE BRISTLE

27

FIGURE 2.8. Circuitry (cf. Fig. 2.7 for key) controlling the identity of mechanosensory (MS) vs. chemosensory (CS) bristles vs.

chordotonal (CT) organs. Those organs are shown with their neurons (N) but without their support cells. They are created from single-cell SOPs (black rectangles) via differentiative mitoses (not shown; cf. Fig. 2.1). In the upper left of each panel are transcriptional states (1= on; 0 = off) of cut and poxn depicted as a binary code. The circuit (middle area) is simple. “Prepattern” genes (cf. Ch. 3) activate atonal (ato) at certain spots in the skin, and Ato suppresses Cut. If the SOP also lacks Paired box-neuro (Poxn), then it makes a CT organ. If Ato is absent, but other proneural proteins (not shown) are present (cf. Ch. 3), then Cut will be expressed and the cell will make a bristle. The bristle type will be CS if poxn is on and MS if poxn is off. The state of poxn (like ato and cut) must be set by region-speciﬁc genes upstream (not shown). Thus, LOF or GOF mutations in poxn interconvert MS and CS bristles, and cut LOF or ato GOF transforms MS bristles into CT organs. Transformations of bristles into olfactory sensilla (not shown) are elicited by amos GOF [1587]. CT organs lack protrusions [378, 1454, 2787, 2831] and arise from the SOP via a different lineage tree [1756]. Stretching of their dendrites (springs attached at two anchor points “A”) causes the neurons to ﬁre an action potential to the CNS [789, 2018]. CT sensilla on the ﬂy femur have 2 neurons [3867], whose cell bodies are embedded along the stretching axis (unlike depicted here). MS bristles transduce touch stimuli by lever action [2174]: deﬂection of the rigid shaft (triangle) toward the body surface (horizontal line) causes it to pivot. Neural depolarization appears to be due to stretching of the dendritic membrane (vs. compression of the tubular body) [789, 874, 2787] (but see [4527]). CS bristles should technically be termed “chemomechanosensory” because they use one neuron to sense touch like MS bristles, and they have four extra neurons whose dendrites extend up the shaft to a pore at the tip [2502, 4125], where they “taste” salts or sugars [111, 2955, 3061, 3529, 4841]. This schematic is adapted from [2174] (MS), [3529] (CS), and [378, 2128] (CT), with logic based on [1455]. Many authors have conjectured that cells differentiate by making a series of binary choices. A favorite metaphor for decision trees has been a rail yard of bifurcating railroad tracks [3063]. N.B.: Expression of cut at the wing margin goes through two phases; only the later phase controls SOP identity [353].

for conditional paralysis (viz., it stops synaptic vesicle recycling) [4759] . Shibire may also play other roles in vesicular trafﬁc [2232, 2900]. Endocytosis typically attenuates signaling by removing activated receptors from the surface [2031] and sending them to lysosomes for destruction [3217] . If true here, then blockage of endocytosis by shi LOF should enhance Notch signaling [702, 3863, 4482] but, as stated above, shi LOF actually causes a N LOF (not a N GOF ) phenotype. This paradox was ﬁnally solved in 2000 by using separate antibodies to follow the fates of N’s extracellular (N-extra) vs. intracellular (N-intra) domains [3271]. It turns out that N-intra cannot be released from the membrane of the “listener” cell until N-extra is engulfed by the “speaker” cell as a ligand-receptor complex. The engulﬁng process requires Shi [3271], and Notch signaling requires release [4156] – hence the similarity of shi LOF and N LOF phenotypes. Because Shi seems to not play a critical role in the listener cell, Numb is probably not acting via Shi in that cell. Notch may be the only ligand that needs endocytosis for transduction per se in discs because other ligands can signal without it [3863]. Nevertheless, endocytosis does appear to propagate certain signals within the epidermis (viz., Dpp [1169, 1546] and Wg [273, 2900, 2984]), perhaps via a “transcytosis” mechanism that uses separate sets of (nontransducing) receptors (cf. Ch. 5) [811, 1154, 2843, 4379, 4694].

Pox neuro and Cut specify bristle type The gene paired box-neuro (a.k.a. pox neuro or poxn) was isolated in a search for DNA-binding motifs of the “paired box” class [401] and was named “neuro” because it is transcribed in neural precursors of the CNS and PNS [950]. In imaginal discs, Poxn is expressed in SOPs of chemosensory (CS), but not mechanosensory (MS), bristles [149, 950]. CS bristles of ﬂies have many features that differ from MS bristles [4125]. For example, they have a shaft that is thin and curved (vs. thick and straight), a characteristic axonal CNS projection, behavioral reactivity to airborne chemicals, and a different lineage that produces 5 neurons (vs. 1) – 4 of whose dendrites extend to the bristle tip (Fig. 2.8) [1741, 1802, 3004, 3061, 3529]. On the dorsal tibia and wing margin, poxn null converts CS into MS bristles [149], while poxn overexpression has the opposite effect [149, 3141, 3142]. Evidently, the role of the wild-type poxn gene is to divert SOPs from a MS to a CS mode of differentiation. One aspect of CS bristles that fails to be transformed in poxn null mutants is the (earlier) times when SOPs arise [149], which may explain the larger bristle sizes (see below). Null mutations at the cut locus convert external sensilla into “chordotonal” (CT) organs in embryos [378, 2831], and a similar transformation (albeit often partial) is seen for adult bristles in cut-null areas of genetic mosaics. Overexpression of cut causes a reciprocal phenotype in embryos [372], although effects in adults have not been

28

reported. CT organs are stretch receptors that fasten to the cuticle (or muscle) at two points and normally lack an external projection (Fig. 2.8) [1454, 2787]. Aside from a sensory neuron (femoral sensilla have 2 [3867]) they have at least 2 support cells: a “cap” cell and a sheath cell whose “scolopale” abuts the neuron’s dendrite [378, 789, 2831]. In this regard, they resemble bristles, which may well have evolved from CT-like structures [378, 3437]. Chordotonal SOPs express atonal (ato) [4898], a bHLH gene located outside the AS-C (at 84F) [2040], but bristle SOPs do not (they express cut instead [149, 2000]). When bristle SOPs are forced to express ato, they stop expressing cut and transform into chordotonal organs [2038]. Evidently, the normal duty of ato is to ensure that SOPs develop into chordotonal organs by suppressing cut (ato cut). SOPs “know” whether to turn on ato due to cis-regulatory enhancers beside the ato transcription unit that respond to region- speciﬁc cues [4208]. Atonal also governs the development of photoreceptor cells (cf. Ch. 7). Based on the normal expression of poxn in cut null embryos [4481] and the normal expression of cut in poxn null discs [149], poxn and cut appear to act independently, despite initial reports (ubiquitous expression and reporter-gene studies) that suggested otherwise (cf. App. 4) [4481]. Apparently, SOPs “read” the state (on or off) of these genes and adopt a particular pathway of differentiation (MS vs. CS vs. CT) accordingly (Fig. 2.8). The failure of cut null to achieve a complete transformation implies that other genes may overlap cut functionally in setting organ-type identity. Candidates include the paired genes BarH1 and BarH2 [3763], which, like cut, are expressed in embryonic external sensilla [1843]. Mutations in these genes switch one type of sensillum to another: deletion of both genes converts “campaniform” to “trichoid” sensilla (see [2128, 2174, 2787] for normal anatomy and [2473, 4309] for transitional intermediates), and overexpression of BarH1 induces the reverse [1843]. Like cut, the Bar genes contain a homeobox [1842, 2286], which suggests that SOPs adopt fates via the same kind of logic that operates on a larger scale through bona ﬁde homeotic genes (cf. Ch. 8) [282]. Olfactory sensilla are governed by yet another bHLH gene named “amos” (absent solo-MD neurons and olfactory sensilla, where MD stands for multiple dendritic) [448, 1587, 1928]. Amos (at 36F) and Atonal (at 84F) are more closely related to each other than to the AS-C family (at 1B) [1756]. Indeed, their basic (DNA-binding)

IMAGINAL DISCS

domains are identical, except for a conservative Arg-toLys change. Nevertheless, they have dissimilar LOF and GOF phenotypes – presumably because different sets of cofactors bind idiosyncratic motifs outside their bHLH domains and steer them to different target genes [1587].

Bract cells are induced by bristle cells The bract is a noninnervated hair made by a single cell [3552, 4531] (Fig. 2.1). It differs from other “trichomes” [2876, 2877, 3362] insofar as it is thick, pigmented, and found only on legs and wings [1361, 1714, 4338]. On the legs, bracts reside next to MS bristles on segments distal to the trochanter [1883]. On the wing, they reside next to MS bristles on the proximal costa [524, 793]. Because the bract cell is not part of the SOP clone [544, 1356, 1800, 4334, 4344], it has long been thought to arise by induction [1808, 3421]. When disc cells are dissociated and reaggregated, isolated bracts are never seen [4334], implying a dependence on bristles. Indeed, bracts do not develop unless both the shaft and socket cells are present [1357, 1808, 3448]: 1. Injecting larvae with Mitomycin C [4335, 4531] or nitrogen mustard [4337] often suppresses bristle sockets, and bracts are missing wherever shafts lack sockets. Likewise, bracts fail to form when sockets are transformed into shafts by N LOF or shi LOF [1803, 3425]. 2. In the mutants Hairless 2 and shaven de (a LOF allele of dPax2 [1319]), bristle shafts are often absent or vestigial [2168], and bracts are missing wherever sockets lack shafts [4338]. Why do some bristles have bracts, while others – only a few cells away – do not? Bristles that normally lack bracts seem to be those that develop earliest (CS bristles and macrochaetes) [149, 1598, 1803, 3142, 3628]. Thus, it is possible that every SOP emits a signal at one stage of its differentiation, but if this stage fails to overlap the period when epidermal cells are responsive, then the bristle will lack a bract. Consistent with this conjecture, CS bristles can acquire bracts when t.s. N LOF or shi LOF mutants are heat-pulsed [1803] (because delayed signals now enter the competence window?), and MS bristles can lose bracts when wild-type pupae are heat-shocked [1803] (because signals are delayed beyond the end of the window?). The Notch pathway is also involved in allocating bracts to particular parts of the wing margin [793, 3092].

CHAPTER TWO. THE BRISTLE

Recruitment of cells by induction is also instrumental in the development of ommatidia in the eye (cf. Ch. 7), olfactory sensilla in the antenna [3531, 3548], and CT organs in the femur [4898]. The latter case is intriguing because it involves iteration (old SOPs inducing new ones) that could theoretically go on forever. In fact, the process is limited by the duration of the competence period and by the extent of the competent region [4898]. The CT inductive signal is transduced by the EGF receptor pathway [4898] (cf. Fig. 6.12). Evidently, this pathway also mediates bract induction because (1) Star LOF causes missing bracts, (2) Star LOF Ras1 LOF double mutants are missing even more bracts, (3) all bracts can be eliminated by heat-treating t.s. Egfr LOF mutants, and (4) ectopic bracts can be induced throughout the epidermis by heat-shocking hs-Ras1 GOF pupae (L. Held, unpub. obs.). Either the receptor (Egfr) or the ligand (unknown) must be localized in accord with proximaldistal cell polarity because bracts are only found on the side of the socket opposite the shaft, and this rule is obeyed even when bristles are misoriented in mutant phenotypes [1357, 1810]. Implementation of the received signal requires the homeobox gene Distal-less (Dll): (1) Dll is expressed in bract, but not bristle, cells on the femur [618]; (2) Dll LOF mutations suppress bracts [618, 4212]; and (3) Dll null clones lack bracts [618, 1561].

Macrochaetes and microchaetes differ in size but not in kind Unlike the sense organs interconverted by mutations in poxn, cut, BarH1, and BarH2, macrochaetes (MCs) and microchaetes (mCs) do not differ in kind. Regardless of size, all tactile bristles share the same cellular composition, lineage, and anatomy. Indeed, a full spectrum of bristle lengths is seen on the legs [1714, 1808, 1883], and mutations can cause notal bristles to have sizes between MCs and mCs [3067, 3373]. Thus, it is not surprising that no “selector” genes have been found that dictate MC vs. mC identity in the same way, for example, that poxn speciﬁes CS vs. MS bristles. Bristle size is a function of polytenization [2475, 4641] – a cyclin-dependent process of DNA replication [961, 1246, 1262, 2551, 4376, 4820] that enlarges the cell’s nucleus and cytoplasm without subsequent cell division [216, 1336, 1764, 3030, 3301] (cf. lepidopteran scales [2338, 4014]). In notal mCs (whose cells are initially diploid), the shaft cell undergoes two rounds of endoreplication, and the socket cell undergoes one [1741]. The degree of polyploidy achieved by MC shaft and socket cells is much greater [2475]. In

29

wild-type males, dorsocentral MCs average 336 µm (posterior) or 250 µm (anterior) in length, while nearby mCs are only ∼80 µm long [3067] and tarsal bristles are ∼50 µm long [1803, 2544]. When SOP mitoses end, the shaft cell is packed with rough endoplasmic reticulum as protein synthesis starts for shaft elongation and cuticle deposition [3216, 3552]. Indeed, ribosomes are probably working at top speed because reducing the supply of any one component slows the entire process. Thus, there are more than 50 sites in the genome where halving the gene dose (i.e., haploinsufﬁciency) causes the same short-and-thin “Minute” bristle phenotype [118, 2561], and the majority of these sites appear to encode ribosomal proteins [3718]. Likewise, bristle size depends on the number of ribosomal RNA genes (at bobbed loci on the X and Y) [1520, 1909, 3604, 4091, 4104]. The situation is analogous to the reduced wing size seen with LOF alleles of rudimentary: this “housekeeping” gene encodes a pyrimidine-synthesizing enzyme, which becomes rate-limiting during wing growth [1314, 1315, 1865, 3838]. Bristle length also depends on two bHLH-PAS genes – spineless and tango – but their role is unclear [1119, 1166]. Larger bristles typically have more longitudinal ridges (up to ∼10) [2474, 3362, 3832, 4639], probably because ridge interval is constrained by an invariant molecular feature such as the diameter of an actin ﬁlament bundle [1645, 1898, 3216, 3351, 3552]. The bundles (up to ∼20 around each shaft [94]) are organized by chickadee [4474], forked [3356], rotundRacGAP [1642], sanpodo [1128], singed [640, 3283], Stubble [94], and tricornered [1430] around the time of cuticle secretion [1642, 3552]. When seen from the side, the grooves in wild-type bristles seem to wind helically around the shaft [3583, 4096], but in fact they meet along a top seam [4639, 4641]. These chevrons suggest that bristles evolved from triangular scales, which rolled up into a cone [1340, 4640]. Shaft differentiation is enforced by dPax2 [1319]: LOF alleles remove shafts, and excess dPax2 elicits two shaftlike spikes from each socket cell [2168]. However, dPax2 probably operates more like a switch because dPax2 GOF can also convert neurons into glial cells [4303]. Bristle cells must divide, endoreplicate, and differentiate before cuticle deposition starts. The larger the bristle, the earlier it should have to begin these events in order to ﬁnish on time. This rationale explains why MC SOPs divide before mC SOPs [1741, 1925], and in general why bristle lengths are inversely correlated with the times of onset of differentiative events (cf. Fig. 3.4c) [806, 2837, 3420]. Similar reasoning explains why SOP mitoses

30

commence earlier for multiply innervated CS bristles than for singly innervated MS bristles within each body region [149, 1598, 1741, 1803, 3142]. They simply need more time to make more cells. If the number of replication endocycles is ratelimiting for bristle size, then delaying the time of SOP initiation should reduce the number of endocycles and cause a smaller bristle. Indeed, tiny bristles are often seen in ﬂies carrying subnormal doses of achaete or scute [801, 802, 4095, 4098, 4108], and the stunting may be due to late onset of differentiation [1462]. Conversely, prodding an SOP to arise prematurely should allow more endocycles and cause a larger bristle. MS bristles that develop from transformed CS SOPs in poxn null ﬂies are probably larger for this reason: the SOPs still arise at an early (unaltered) stage and hence have more time than neighboring MS SOPs to endoreplicate. SOP initiation can also apparently be accelerated by augmenting AS-C gene function [1563, 2690]: 1. Notal mCs can become MCs when lethal at scute is overexpressed (via a UAS-l’sc transgene driven by scabrous-Gal4) [3027, 3037]. 2. Derepression of the AS-C in the dorsocentral notum (in polychaetoid LOF mutants [733]; Ch. 3) changes mCs into MCs [3067], while the opposite transformation (MCs into mCs) is seen when spalt (an AS-C regulator) is overexpressed [987]. 3. When combined with mutations that derepress the AS-C, ac GOF (Hairy wing) alleles transform certain leg bristles into MCs [1802]. 4. Ectopic overexpression of sc in the eye can convert interommatidial mCs to MCs [4209]. 5. Extra doses of sc (in a background where sc is dere-

IMAGINAL DISCS

pressed) convert tergital mCs to MCs [2690], and similar enlargements occur when sc is overexpressed using a heat-shock promoter [3628]. (See [3905] for how wingless may be setting AS-C levels for this purpose.) When cells are forced to divide by artiﬁcially expressing the mitosis-licensing agent String, unusually small bristles often arise at sites where SOPs are being recruited [2225]. This effect has been attributed to a delay in SOP initiation (consistent with the above argument), rather than to precocious entry into the SOP differentiation pathway [2225]. In summary, it appears that bristle size is controlled by rheostats, rather than by switches. A bizarre bristle transformation occurs in ﬂies with a construct of the male-speciﬁc cDNA of doublesex (a gene involved in sex determination) driven by a heatshock promoter. Whereas most leg bristles in wild-type ﬂies are thin, tapered, and brown, most leg bristles in these ﬂies are thick, blunt, and black, like bristles of the sex comb [2106]. Whether this trait involves changes in the timing of SOP events remains to be determined. Evolutionarily, bristle development has been entrained by various anatomic gradients. For example, bristle lengths increase from anterior to posterior among notal MCs [805, 2837, 3067], tergital mCs [801, 2690], and sternital bristles [804]. They increase from dorsal to ventral for tergital MCs [801], from ventral to dorsal for tarsal bristles [1803], and from distal to proximal for leg bristles in general [1714, 1883]. Bristle intervals typically increase with bristle lengths. Various aspects of bristle patterns have also been tailored to ﬁt certain features of the cuticular landscape. Explanations for some of these correlations are presented in Chapter 3.

CHAPTER THREE

Bristle Patterns

The epidermis of a D. melanogaster adult has on the order of 500,000 cells, ∼5,000 of which (∼1%) make bristles [1804]. A priori, it would seem reasonable to expect bristles to sprout as randomly as the hairs on a human arm. However, even the most scattered bristles – the tergite microchaetes (mCs) – have fairly uniform spacing [801, 2301--2303]. At the other extreme of precision are the 40 macrochaetes (MCs) on the head and thorax, whose basic layout has been conserved for 50 million years (Fig. 3.1) [1625, 3966, 3980, 4185]. Except for the MCs, the bristles of each body region tend to vary in number and position from one ﬂy to the next. Interestingly, most bristles are organized in rows that run parallel or perpendicular to axes of the body or limbs [1808, 2883, 4015]. Within such rows, the bristles are aligned more or less accurately and are spaced more or less evenly. Different rules govern different patterns. Thus, notal mCs form jagged rows along the anteriorposterior axis [805, 4428], while wing bristles form straight rows along the margin [1741], eye bristles arise at alternating vertices of each ommatidium [3539], and belly (sternital) bristles are spaced at intervals proportional to their shaft lengths [804]. Why do such patterns exist? Surely, some are adaptive. For example, ﬂies use “brushes” (parallel transverse rows) on the legs to wipe dust from their eyes [4225], and other patterns appear to map air currents, prevent wetting, or act as shock absorbers [1116, 3206]. However, many may simply be accidents of evolution [3966]. Algorithms that evolved to align neuroblasts in the embryo’s central nervous system (CNS) may later have acquired the duty of building an adult peripheral nervous system (PNS) [3981, 3984]. Old tendencies to obey particular cues may

have been retained in the new disc-speciﬁc contexts [1804, 1807], causing bristles to sprout at certain spots, interfaces, or contour lines (cf. Chs. 5 and 6). Moreover, bristle patterns in different species may manifest whimsical idiosyncrasies. Variations within the genus [376, 1361, 1801] imply a frivolity that rivals the pastiches of colored scales on butterﬂy wings [210, 434, 3119]. Regardless of why bristle patterns arose, the designs themselves epitomize the general problem of pattern formation in development. Namely, what causes the correlation of particular cell types with particular positions? We are far from a complete explanation, but it is now clear that the diversity of the patterns is deceptive. In fact, all of them use the same genetic tools to enable each epidermal cell to make the Hamletian choice of whether to be, or not to be, a bristle. Given the two dimensionality of the cuticle and the single-cell origin of each bristle, the riddle posed by all these patterns is reducible to a simple Euclidean problem. To wit, what kind of geometry does the genome use to place points in a plane?” Throughout the twentieth century, many of the intellectual giants of genetics tried to solve this “Bristle Plotting Puzzle” by clever experimentation or deft theorizing. Most of the speculation focused on mutants with abnormal numbers of MCs. T. H. Morgan toyed with the genetics of extra-bristle mutants as early as 1915 [2640]. Morgan’s student, A. H. Sturtevant, was more intrigued with missing-bristle mutants, whose asymmetric patterns are as eye-catching as a gap-toothed smile. In 1918, he used Dichaete in artiﬁcial selection trials [4178]; in 1931, he toppled a popular theory about scute [4187]; and in 1970 (just before his death), he showed a dose

31

32

IMAGINAL DISCS

Lattice Tandem Regular Row Irregular Row Constellation Isotropic

FIGURE 3.1. Diversity of bristle patterns on the ﬂy surface.

In this adult male, the eye bristles occupy alternating vertices of ommatidia in a hexagonal lattice. The sex comb typiﬁes a “tandem” array where bristles touch. Other leg bristles are aligned in straight (“regular”) rows and spaced evenly. On the notum the rows tend to be irregular, especially laterally. The four scutellar macrochaetes form an aperiodic (“constellation”) pattern that is constant from one ﬂy to the next, as are the arrangements of other macrochaetes. Most tergite bristles (except those at the posterior edges) are arranged randomly (“isotropically”) but spaced uniformly. Tergite pigmentation has been omitted for clarity. Adults are ∼3 mm long [3421]. Adapted from [1804]. For anatomical nomenclature, see [1224]. See also App. 7. What sort of geometry does the ﬂy use to create these patterns? Do all the arrays arise via the same algorithm? Do bristles know where they are? The mysteries of this “golden ﬂeece” have intrigued investigators for decades [1799, 1804, 1808, 2434, 4096].

dependency of achaete and hairy that later proved prescient [4185]. Another famous member of the Morgan lab, H. J. Muller, inspired a team of Russian researchers to dissect the achaete-scute Complex (AS-C) [650] – an enterprise that later became a cottage industry for Antonio Garc´ıa-Bellido and his school in Madrid [1354, 1452]. British patriarchs John Maynard Smith [2768, 2769] and C. H. Waddington [4517] invented cute schemes for categorizing mutant patterns, while Vincent Wigglesworth and Peter Lawrence speculated about the chaotic origins of natural patterns [2427, 2439, 4657]. Lawrence and Garc´ıa-Bellido used bristle patterns as a platform for

digging into deeper issues of development, and these two mentors trained many talented students who became leaders in this ﬁeld. Indeed, bristle pattern research is so interesting from a historical standpoint because of these crosscurrents. Its parochial models have often inﬂuenced the larger ﬁelds of development and genetics and vice versa. For instance, Curt Stern’s ideas about bristle patterning (discussed below) colored Lewis Wolpert’s thinking about patterning in general [4724], and Wolpert’s theory of positional information in turn revolutionized the bristle subﬁeld via the team founded by Howard Schneiderman

CHAPTER THREE. BRISTLE PATTERNS

and Peter Bryant in California [530, 1485]. Most of the old questions have now been answered by modern molecular techniques, but the story of how we got here is as charming as the answers themselves.

Surprisingly, different macrochaete sites use different signals Alan Turing, the British mathematician who cracked Germany’s Enigma Code in World War II [1863], is also famous for helping to found the modern ﬁeld of computer science [872, 4409]. It is less widely known that Turing also dabbled in developmental biology. In 1952, he proposed a new theory to explain how biological patterns arise [4410]. He argued that chemical reactions among randomly diffusing substrates should cause the products to accumulate at regular intervals. Given the right parameters, his formulae predict stable spots or stripes whose spacing depends on diffusion constants, and whose arrangement depends on the shape of the region where the reactions occur [1176, 2769, 3014]. Turing’s ideas founded a ﬂourishing school of thought about the abilities of reaction-diffusion systems to create patterns [1727, 2806, 3013]. In 1954, Curt Stern invoked comparable dynamic forces to explain how bristle patterns develop [4095, 4096]. Stern (1902–1981) had learned ﬂy lore as a postdoc in Morgan’s lab [2614, 3070], where he earned kudos for proving that crossing over occurs in somatic tissues [119, 4089, 4347]. Poetically, this very phenomenon allowed him to test his new ideas [4100, 4103]. He acknowledged that chemical interactions such as those in Turing’s Model could mediate patterning, but he thought that physical forces might work just as well. For example, Stern conjectured that uneven growth in a sheet of cells might lead to stress points, which cells could then “read” as signals for making bristles. The initiation of differentiation of specifically localized organs cannot be a purely locally caused process but is possible only on the basis of a prepattern which itself is the result of interrelation of parts of a developing whole. [4098] Possibly bristles differentiate at places of specific strains or stresses set up in the folded embryonic tissue which later smoothly covers the thorax. . . . Of course, the physical scheme of tensions used in this train of thought is only one of many variants. Different general patterns of biochemical properties . . . in the embryonic future epidermis could be considered equally well. [4096]

Stern coined the term “prepattern” for the array of signals that preﬁgures the ﬁnal pattern. Initially, he imagined that the sites containing the cues must be in-

33

terdependent. He called those sites “singularities.” In the above example, the singularities were the stress points, but in general they could be any heterogeneity. This idea came to be known as the “Prepattern Hypothesis” [4346]. In a population of somatic cells which are genetically alike, the differential fate of some specially located ones must be due to a superimposed differential organization of the population. . . . The existence of a differential organization, a ‘‘prepattern,’’ within any imaginal disc must precede and hence be responsible for the patterned origin of bristles. [4095]

Because the prepattern was deemed to be a gestalt, any change in one of its components should alter the remaining components in the same way that tugging one node of a spider’s web will alter the whole web. To test whether inﬂuences spread beyond the point of an interference, Stern used mosaic ﬂies with discrete regions of mutant and wild-type tissue. He created the genetic patchiness by somatic crossing over. His reasoning was straightforward. If a mutant ﬂy differs from the wild-type in its network of force vectors, then mutant tissue in a wild-type background should cause novel webs of interactions, and strange new patterns should emerge that might yield clues to the mechanism. Over the next ∼20 years, Stern and his co-workers tested this “Force Field Model” by analyzing ∼20 possible “prepattern” genes in mosaic settings. Disappointingly, they found virtually no new patterns [4100, 4346]. Nevertheless, some clues were uncovered that led to profound insights. Ironically, the pithiest clues came from one of the ﬁrst mutations that Stern studied – ac1 , an allele of achaete (ac), whose name means “no bristle” in Latin. This gene was mentioned earlier as part of the AS-C (Fig. 2.4). Homozygous ac1 ﬂies usually lack a notal MC called the “posterior dorsocentral” (PDC). In mosaics, this PDC bristle is typically absent if its site is occupied by ac1 tissue and typically present if heterozygous tissue resides there [4095, 4096]. Evidently, cells at the PDC site use their own genotype to decide whether to make a bristle, regardless of the distribution of genotypes elsewhere in the disc. This local autonomy was not what Stern expected based on his Force Field Model, nor was it what Turing’s Model would have predicted since diffusing chemicals should naturally affect nearby cells [2769, 4016]. Given the ability of ac1 tissue to suppress the PDC bristle but virtually no other MCs when located at extraneous sites, Stern surmised [4100] that

34

IMAGINAL DISCS

Prepattern

Pattern

Force Field Model Filter Model Prepattern A

Competence Signals

Receptors A B C D E F

B

C D E H

H I J K L

F G J

I Filter

K L

Bristle Differentiation

CHAPTER THREE. BRISTLE PATTERNS

1. ac1 must interfere with the “competence” of cells to respond to a speciﬁc signal at the PDC site. 2. Different MCs must use different signals. His second deduction is a special case of the general Principle of Nonequivalence [2516, 4727, 4729] – the idea that identical structures are speciﬁed by nonidentical signals. This notion is counterintuitive because a priori Occam’s Razor suggests a minimum number of signals for any given pattern. In a similar way, stripes of pair-rule gene expression in the embryo blastoderm were once thought to arise via spatially periodic signals in a Turinglike process [1728, 2159, 2377, 2379, 3031]. However, there, too, the Principle of Nonequivalence trumped Occam’s Razor: most stripes are actually governed by unique combinations of transcription factors [104, 1908, 3247, 3587, 4834]. Thus, MCs – like pair-rule stripes – are made “inelegantly” [49, 2378, 2765, 3248]. Based on the above conclusions, Stern revised his model by inserting a new step (Fig. 3.2). For a prepattern signal to evoke a pattern element in the ﬁnal array, this signal must ﬁrst be transduced by the cell at that site. To illustrate this “competence” step, he used a prospecting analogy where signals differ qualitatively from one another: This then is a picture of gene action in differentiation. A whole developing embryo, or a developing part of an embryo, possesses regional differences of varied nature. One gene (or its products) responds to one specific regional peculiarity, another gene responds to other regional characteristics. These responses, which are the necessary prerequisites for a complete series of processes, result in specific differentiations. One may compare the groups of genes present in all regions

35

of the developing organism to groups of prospectors working in different regions of a country. Each group would be alike in consisting of different men looking for coal, gold, uranium, and so on. In some regions only the coal prospectors would respond to the coal deposits, at other places the gold prospectors would respond to gold veins, at still other places the uranium prospectors to ores of this metal. Each response would depend on the pre-pattern of geological differentiation. Each response would be the necessary prerequisite for subsequent differentiation in the form of the specific structures erected by mining engineers -- perhaps deep shafts and elevators for coal mines, surface excavations and washing troughs for gold, shallow tunnels and radiosensitive installations for uranium. [4096]

In this new formulation, mutant genes that affect the pattern should be classiﬁable into those that do so by altering the prepattern (“prepattern genes”) and those that do so by modifying cellular competence (“competence genes”). Signals that cannot be transduced due to a competence mutation like ac1 would be “ﬁltered out” before they elicit bristles at those sites [4100]. One way to envision this “Filter Model” is with cell surface receptors serving as the ﬁlter (Fig. 3.2), although Stern never advocated any particular molecular device. If each bristle site has a unique signal and each signal has a unique receptor, then the failure of ac1 cells to make a PDC bristle could be due to a defective receptor. In fact, the ac gene does not encode a surface receptor, nor does the ac1 mutation disable the Ac protein. Rather, ac encodes a nuclear protein, and ac1 deletes a region-speciﬁc cis-enhancer that lies 5 kb upstream from the ac gene’s transcription start site. As explained below, it is the ensemble of these enhancers at the AS-C locus that actually “ﬁlters” the prepattern signals.

FIGURE 3.2. Curt Stern’s Prepattern Hypothesis in its original (above) and ﬁnal (below) forms, both of which were proposed in

1954 [4095, 4096]. Force Field Model. Initially, Stern envisioned patterns as the steady-state outcomes of dynamic physical or chemical forces. In this illustration, uneven growth deforms the epithelium so that each half mesonotum acquires 7 stress lines. Each of the 11 sites (“singularities”) where the lines intersect then induces a MC. Because mutations can only change the pattern by altering the force vectors, any patches of mutant tissue in a wild-type background should distort the entire web of forces. Contrary to expectation, virtually every pattern-altering mutation that Stern studied (including achaete1, Fig. 3.3c) affected differentiation autonomously in mosaics. Filter Model. To explain why cells exhibit such insularity, Stern postulated that (1) each MC must use a different signal (A–L, although signals could differ quantitatively instead), and (2) a MC will only form at a site if the cell there is “competent” to respond to that speciﬁc signal. Inserted between the prepattern and the pattern is an extra step wherein each cell “ﬁlters out” signals that it cannot “hear.” In this illustration, the competence of each cell (depicted with apical microvilli) depends on surface receptors. In reality, positional signals are “ﬁltered” through enhancer elements within the achaete-scute Complex (Fig. 3.4) [2891]. Wild-type cells (like this one) cannot respond to Signal G, a “cryptic singularity” that Stern discovered [4097]. Similarly, ac1 cells would be “deaf” to Signal I (not shown). N.B.: Stern described these variations of his theme in general terms, but never formalized or named them as speciﬁc models. Here they are made explicit to highlight their differences. The idea that tensions affect gene expression (top model) seemed farfetched at the time, but various examples are now known [301, 761, 762, 1942, 2580].

36

Prepatterns may contain hidden ‘‘singularities’’ In male ﬂies, the foreleg bears a “sex comb” [1716], socalled because its single ﬁle of stout bristles resembles a hair comb [4344]. In gynandromorphs, sex comb “teeth” can arise when even a small patch of male tissue (in an otherwise female leg) overlaps this site [3441, 4105]. Evidently, female forelegs have inductive signals for teeth but lack a comb because female cells are not attuned to them. This case is interesting because it shows that even wild-type cells routinely “ﬁlter out” certain signals in their own epidermis. Extending this reasoning into the realm of evolution, Stern argued that signals may arise evolutionarily before the ability of cells to respond to them: Many Drosophila species are known which do not possess a sex comb in either sex. D. melanogaster undoubtedly evolved from such sex-combless ancestors. Must we postulate that a basically new configuration of a foreleg had to originate in evolution before a sex comb made its appearance; or can we not more simply assume that the sex comb pre-pattern was long in existence in the legs of flies and had to wait only for a mutation toward a single new gene which could respond to the hitherto unheard call? [4096]

In other words, ancestral drosophilids may have possessed a signal for sex combs, but this signal failed to evoke comb development until D. melanogaster males evolved the ability to “hear” it. Stern called such hidden signals “cryptic singularities” [4100], and he uncovered another instance in studying a certain duplication of the yellow gene, where AS-C enhancers reside. D. melanogaster normally lack a MC at the “interalar” site [3966, 4185] (“G” in Fig. 3.2) on the notum (dorsal thorax). However, an extra MC can arise there when the site is occupied by duplication-bearing cells, even if the rest of the notum is wild-type. Apparently, the duplication enables cells to “read” a subliminal signal that wild-type cells cannot [4097]. Because other (ancestral?) dipterans have an interalar bristle, D. melanogaster may have retained a vestige of the signal but lost the ability to transduce it, in which case any mutation that reveals it would be considered atavistic [1354]. Other authors have adduced evidence for cryptic singularities in line with the dorsocentrals [2640, 4507], between the ocellars and orbitals [2769, 4013], between the anterior and posterior scutellars [1282, 2413, 3890, 4517], and on the tarsus [1802]. Indeed, subthreshold bristle-forming tendencies have been found in nests of cells that nor-

IMAGINAL DISCS

mally do not form a bristle, but can be prodded to do so in certain mutant backgrounds [589]. There are two ways of thinking about singularities in general, regardless of whether they are overt or covert: 1. Singularities are instructive signals, tantamount to telling a cell to “Make a bristle!” 2. Singularities are not signals per se, nor are they inherently tied to any particular structure. Stern favored the second view [4098], as illustrated by the following passage where he cites some classic experiments of Spemann’s [4030]. There are some famous experiments in which embryonic tissues of a salamander have been transplanted into the mouth region of frog embryos, and reversely frog tissue transplanted into the mouth region of salamander embryos. These experiments have shown that both salamander and frog are endowed with a prepattern which distinguishes their mouth regions from the rest of the body. . . . In the evolution of the two branches of amphibians to which salamanders and frogs belong, no more than the appearance of new genes which responded in different ways to the previously established inductive capacity of the mouth region was required. The complexities of any organism must imply an inexhaustible array of prepatterns, a few of which are realized in accomplished differentiations and most of which are waiting for the response of genes not yet arisen in a specific organism. [4096]

The molecular-genetic analyses described below support this argument. Singularities appear to be transacting factors that elicit bristles only because the AS-C has cis-enhancers to bind them. In principle, any bristleless site could be a “cryptic singularity,” because evolution could endow the AS-C with enhancers to “read” virtually any position-speciﬁc protein (cf. Ch. 8) [662]. Thus, each prepattern is an accidental result of the cross-linking (by evolution) of bristle differentiation to certain heterogeneities of gene expression in the epidermis. Our understanding of prepattern mechanisms rests mainly on studies in the wing disc, where many transacting factors are known. Those agents will be examined in Ch. 6 after readers have been briefed on the key signaling pathways (cf. Table 6.2 and Fig. 6.14). The rest of this chapter focuses on the AS-C itself and its regulation by circuits that operate wherever bristles arise.

How Achaete and Scute control bristles was debated for decades Among ac1 mosaics that lack a PDC due to homozygous ac1 tissue occupying that site, a MC often arises

CHAPTER THREE. BRISTLE PATTERNS

nearby in heterozygous territory (Fig. 3.3) near the normal site [802, 3613, 4095, 4098]. This apparent ability of the PDC to be displaced led Stern to infer that the PDC can come from any cell within a cluster of equivalent cells at this spot. When the cell at the “preferred” site within this “equivalence group” is prevented from becoming a MC, another cell can substitute. Why don’t all of these cells make MCs? Stern guessed that whichever cell emerges as the SOP must block the others from adopting this fate. He likened the situation to embryonic ﬁelds, where the tissue area that can potentially make a structure (e.g., a limb [1955, 4589]) is typically larger than the region that actually does so. This astonishing result fits perfectly well into existing concepts of the embryologist. He has discovered the existence of pre-patterns which he calls embryonic fields. These are areas in which a specific differentiation may occur anywhere. Actually, under normal circumstances, the differentiation takes place in only a limited part of the whole field, at a peak, figuratively speaking. Once differentiation has set in at the peak, no other differentiation occurs within the larger field. If, however, differentiation at the peak is suppressed, then a lower region of the field may differentiate. This differentiation itself will exclude other differentiation within the remainder of the field. . . . The pre-patterns of the embryonic tissue of Drosophila, which call forth the response of genes involving the differentiation of bristles, are embryonic fields of larger dimensions than the limited points of normal location of bristles. If the influence of the achaete gene at the normal point prevents differentiation of a bristle, other parts of the field may assume the properties of peaks and differentiate bristles. [4096]

Given the directionality of the displacements, Stern deduced that the PDC cluster must be elongated along the anterior–posterior axis, rather than having the circular shape expected if the sensory organ precursor (SOP) emits a diffusible inhibitor.

37

tant ﬂies with null alleles for both genes. Such ﬂies lack virtually all bristles due to a failure of SOP inception [912] but otherwise look normal [1379, 1462, 1802, 3812]. Not only are ac and sc necessary, they are also sufﬁcient for bristle initiation because overexpression of either gene can induce extra bristles at ectopic sites [191, 635, 1348, 3628] by evoking extra SOPs [912, 3982]. AS-C polymorphisms account for much of the naturally occurring variation in bristle number seen in this species [2593, 2645]. Historically, this gene complex was viewed as holding the key to deciphering bristle patterning [1354, 4766], and rightly so. Partial-LOF sc mutants are striking because they create holes in an otherwise perfect pattern of MCs, and because the bristles that they remove depend strictly on the allele. Among the 11 notal MCs {PSC, ASC, PPA, APA, PDC, ADC, PSA, ASA, PNP, ANP, PS} (Fig. 3.4), the subsets that tend to be missing include sc1 {PSC, ASC}, sc4 {PSC, ASC, PPA, APA, ASA, ANP, PS}, sc6 {ANP}, sc7 {PSC, ASC, PPA, APA}, and sc9 {PSC, ASC, ANP} [1360]. How does the AS-C encode bristle positions? This “Bristle Coding Enigma” bafﬂed geneticists from the discovery of the ﬁrst sc LOF allele (sc1 ) by Calvin Bridges in 1916 [470] until 1995 when the mystery was solved by Juan Modolell’s team in Madrid [1538, 2890]. The intervening 80 years were punctuated by intriguing clues and colorful controversies [628, 650, 1354, 1799, 2891]. Some milestones in this saga are recounted below. 1916

It seems that differentiation of a typically located [PDC] inhibits other nearby potential bristle differentiations in the line defined by the position of the dorsocentral bristles but that absence of the [PDC] permits the realization of some of these potencies. [4095]

The inhibitory mechanism is considered later. It does not appear to depend on achaete or any other genes in the AS-C. The AS-C spans ∼100 kb near the tip of the X chromosome (Fig. 3.4). Four of its genes – achaete, scute, lethal-at-scute, asense – are termed “proneural” because they promote neural development in the CNS or PNS [1454, 3637]. Innervated bristles depend almost exclusively on achaete (ac) and scute (sc), as shown by doubly mu-

1920s

On January 22, the ﬁrst scute mutant was found by Calvin Bridges [470]. Twelve days later (Feb. 3), the ﬁrst achaete mutant was found by Alexander Weinstein. Both men were then graduate students in Morgan’s lab [2283], where the Great Quest for mutants had begun in 1910. Neither achaete nor scute could be given a one-letter abbreviation because “a” and “s” had been taken by arc and sable (in 1912) [470]. Hence, the two genes became known as ac and sc. Following H. J. Muller’s ﬁrst sojourn in the Soviet Union, A. S. Serebrovsky used Muller’s X-ray method to induce new sc mutations [650, 651]. His assistant, N. P. Dubinin, found partial complementation among alleles [650] and used these data to deduce a linear array of sc “subgenes.” Each subgene was supposed to control a discrete subset of bristles [1108, 1109]. His inferred sequence of subgenes

38

IMAGINAL DISCS

wild-type

achaete1 (y null)

mosaics

PDC

a

b

Proneural Cluster

c

e

d

f

Proneural Cluster (Remnant)

Proneural Cluster Halo PDC

g

1931

h agrees roughly with what we now call “enhancers” along the AS-C [1360, 1453]. A similar mistake was made at the Bithorax Complex [2507, 2782], where enhancers were thought to be genes per se before DNA cloning clariﬁed the situation [146, 1121, 2672, 4671]. A. H. Sturtevant and Jack Schultz supported the “linear array” tenet of Dubinin’s “Sub-

ADC

ADC

PDC

PDC

i

j gene Hypothesis” but disproved the “discrete subset” tenet by showing that allelespeciﬁc bristle groups can be modiﬁed via changes in the genetic background [4187]. This malleability implied that the patterning of bristles is “a system in which limiting factors are important and in which, therefore, changes in the rest of the system make

CHAPTER THREE. BRISTLE PATTERNS

39

FIGURE 3.3. Proneural clusters, whose existence was deduced by Stern in 1954

[4095]

and conﬁrmed histologically in 1989–91

[912, 3637, 3982].

a–f. Stern’s reasoning, based on the nonautonomous formation of an ectopic MC by wild-type tissue in achaete1 (ac1 ) mosaics. a. Heminotum (minus humerus) of a wild-type ﬂy, with all bristles omitted except the PDC MC. b. Homozygous ac1 ﬂies typically lack a PDC. Absence of shading indicates yellow null (ynull , yellow cuticle), which was a linked marker for ac1 1 1 null ac1 sn1 /+++ [636, 1442], as was singed (sn , gnarled bristles). c, d. Gynandromorphs arising from X-chromosome loss in y heterozygotes (redrawn from [4095]). Both heminota lack a PDC at its normal site, evidently because the tissue there is mutant. In some cases (e.g., d), a PDC-like MC arises in nearby heterozygous tissue [3613, 4095, 4096]. e, f. Illustrations based on Stern’s hypothesis for bristle displacement [4095]. Squares show imaginary details for PDC areas in a and d. Dashed line marks the normal PDC site. A “proneural cluster” (dark shading) of cells (small hexagons) is “competent” to make a PDC. The central cell becomes the PDC (e) unless its ac1 genotype renders it unable to do so (f), whereupon a cell in the remaining part of the cluster substitutes, thus displacing the PDC. All SOPs, Stern argued, must inhibit their neighbors from making bristles. g–j. Drawing of a real cluster on a right heminotum at ∼25, 20, 10, and 0 h before pupariation, as revealed by antibodies to Scute [912, 3982]. Ovals are nuclei (diameters ≈ 4 µm). The actual dorsocentral cluster (1) has an irregular shape, (2) grows, and (3) yields two MCs whose SOPs arise at the cluster’s edge. g. Some cells express more Scute (dark shading). h. The SOP for the PDC (black) arises in a high-Scute area and acquires a “halo” of low-Scute cells. i. As the cluster grows, the high-Scute area shifts and the ADC (no halo yet) arises. The reason for this shift is not known. j. In this cluster, non-SOP cells stop expressing Scute before SOPs, which do so later (before dividing). Further growth widens the gap between the ADC and PDC. As for why the ADC and PDC typically form on the medial (dorsal) side of their common cluster, the answer appears to be that a diffusible bristle-promoting signal (Dpp) encounters this cluster on its medial face and then wanes (cf. Fig. 6.14) [3373, 4368].

1932

1935

changes in threshold values.” Dosage thresholds have since been shown to be decisive in bristle patterning and to depend on a balance between opposing groups of HLH genes (see below). Sturtevant and Schultz advocated a version of Plunkett’s “Spreading Model” [3405], wherein a dorsocentral source (“center”) emits a bristle-inducing substance, and different sc alleles impede its diffusion in different directions by acting at different times. In the same year, Richard Goldschmidt (Curt Stern’s colorful mentor [4102]) proposed a similar diffusion-based theory [1520]. Muller [2986] and Sturtevant [4181] independently disproved the Spreading Model by demonstrating the same autonomy for sc that Stern later showed for ac [4095, 4096]. To wit, they showed that scLOF tissue can remove bristles without affecting nearby wildtype tissue – a result that is inconsistent with the involvement of diffusible factors. Our modern view of AS-C function afﬁrms Muller’s conclusion that “the development of bristles . . . is not governed by one or a few centers, but is in its major features autonomous at the site of each bristle.” Like Sturtevant and Schultz, George Child disproved Dubinin’s “one-subgene-onebristle-group” idea but did so by manipulat-

1954

1975

1978

ing the environment instead of the genetic background. He showed that allele-speciﬁc traits can be changed by raising scLOF ﬂies at different temperatures [767--769]. Curt Stern proposed his Prepattern Hypothesis [4095, 4096]. The autonomy that he found in ac mosaics ruled out a Force Field Model for prepatterns. He concluded that cells respond to position-speciﬁc, bristle-inducing signals only if they are “competent” to receive them. Displacements of wild-type bristles led him to postulate that (1) each MC arises within a group of equipotent cells, any one of which can become a bristle; and (2) the nascent SOP inhibits its neighbors from adopting a bristle fate. Clifton Poodry used irradiation to probe the timing of bristle differentiation on the notum [3420]. He found that when MCs are grouped according to the time when they acquire insensitivity to irradiation, the groups do not match the subsets removed by scLOF alleles. Thus, those subsets cannot be explained by loss of sc function at allele-speciﬁc times (as Sturtevant and Schultz had imagined). The time when a bristle site becomes insensitive to irradiation was found to be correlated with bristle length [2837]. This trend is probably due to a need for larger bristles to begin differentiating earlier in order to

40

IMAGINAL DISCS

Macrochaetes

a SC PA DC SA NP PS HU VT PV OC OR P A P A P A P A P A D V P A P M A

70 kb

60

y

82

50

40

20

10

0

l’sc sc AS-C

ac

64

30

-10

-30 kb

-20

ase T1

T2

ac1

b SC PA DC SA NP PS HU VT PV OC OR P A P A P A P A P A D V P A P M A

0

1

2

3

4

5

6

7

8

9

10 11 12 13 14 15

Onset of radio-insensitivity (h AP)

Bristle length (arb. units)

c

80 pSC aPA 70 pDC

aSC

60 pSA 50

aNP aVT OC

40

aDC dHU

PS pPA aOR PV

30 20

0

1

2

3

4

5

6

7

8

9

10 11 12 13 14

Onset of radio-insensitivity (h AP)

1985

ﬁnish before cuticle deposition. This discovery (Fig. 3.4c) had no bearing on why different scLOF alleles affect speciﬁc bristles. Cloning of the AS-C was reported by Juan Modolell’s team [636], which had previously

cloned the sc portion [656]. Most sc mutations were found to map within a 50-kb region of DNA proximal to the sc gene. Nucleotide sequencing revealed similarities (and redundancies) of the four bHLH genes in the

CHAPTER THREE. BRISTLE PATTERNS

41

FIGURE 3.4. Spatiotemporal control of macrochaete development on the head and thorax.

a. The genotype-phenotype “switchboard” for MCs. Half head and thorax of a wild-type ﬂy (dorsal view; mCs omitted). Black circles below symbolize MCs seriated in posterior-to-anterior order, except that some are grouped (cf. Fig. 2.6 for key: P = posterior, A = anterior, D = dorsal, V = ventral, M = middle). The double helix spans the ∼100-kb achaete-scute Complex (AS-C) near the tip of the X chromosome. Arrows denote transcription units (kink = intron), with distal to the left and zero as an arbitrary EcoR1 site. Four “proneural” genes (achaete, ac; scute, sc; lethal-at-scute, l’sc; asense, ase) encode bHLH proteins, although only ac and sc normally affect MCs. Enhancers (rectangles) were mapped genetically (light shading; double arrows mark unknown limits) [636, 3688] or by their ability to drive reporter genes in proneural clusters (dark shading) [1538, 2891]. The minimal effective sequence per rectangle is unknown except for the DC enhancer [1380]. Note that the MCs and enhancers are not colinear (dotted lines). Genetic loci for ANP and APA enhancers (5–7 kb and −1 to −7 kb, respectively; not shown) differ trivially (a few kb) from those depicted (as assessed by reporters). The ac1 allele affects DCs (Fig. 3.3) due to an 18-kb (64–82) deletion that overlaps the DC enhancer. The functional core of this 5.7-kb enhancer is really only ≤1.4 kb long [1380]. The DC enhancer may be looped into contact with the sc promoter (30 kb away) when Pannier (a transcription factor) binds this enhancer and forms a complex with bHLH dimers that bind the promoter [3504]. The yellow gene (y) has its own enhancers (not shown) [1442, 1534]. In Ref. [1538], ASA and PPA are (mistakenly?) ascribed to the same enhancer. See also App. 7. b. Times (hours after pupariation) when sites become insensitive to γ-irradiation. Because the high dose that was used (104 Rad) kills dividing cells but spares postmitotic ones, onset of insensitivity probably signiﬁes completion of SOP mitoses [3420]. Indeed, onset times agree roughly with histological detection of four SOP progeny at these various sites [1925]. More precisely, times may reﬂect the start of endoreplication because shaft cells become insensitive and endoreplicate ∼1–3 h before socket cells [1741, 3420, 3421]. No colinearity exists between onset times and the overall posterior-anterior seriation (dotted lines), although posterior bristles precede anterior ones within most pairs – a correlation that may have less to do with location than with shaft length (see c). Times for mCs are 19–24 h AP (not plotted). c. Inverse correlation between bristle length and onset of radio-insensitivity. This trend may stem from a need for SOPs to begin mitoses or endoreplication early enough to ﬁnish differentiating before cuticle deposition. Strangely, the earlier a MC SOP arises (cf. Fig. 3.7), the longer it waits before starting mitosis [1925]. Lengths (arbitrary units) are for wild-type ﬂies raised at 17◦ C [2837], times are from [3420], and the line was plotted by best-ﬁt regression. N.B.: For calibration purposes, the actual lengths for pDC and aDC bristles are 336 µm and 250 µm, respectively [2544].

1987

1989

1991

complex [73, 258, 4485]. Surprises included the paucity of transcription units (only 6 genes in 100 kb of DNA) and the absence of introns [2891]. Ruiz-G´omez and Modolell used 74 terminal deﬁciencies with breakpoints in the AS-C for a deletion analysis [3688] and tested two hypotheses: (1) the “Differential Threshold Model,” where each MC needs a different amount of Ac or Sc, thus explaining why bristles are affected in a deﬁnite sequence as enhancers are whittled away; and (2) the “Site-speciﬁc Enhancer Model,” where each enhancer controls sc expression at a certain site. They favored the latter model, which turned out to be correct. The groups of equipotent cells that Stern predicted in 1954 were documented histologically by in situ RNA hybridization and named “proneural clusters” (PNCs) [1454, 3637, 3957]. In situ detection of Ac and Sc protein distributions allowed PNCs to be resolved in greater detail [912, 3982]. James Skeath and Sean

1992

1995

Carroll validated Stern’s fanciful idea of cryptic singularities by identifying an “inactive cluster” anterior to the dorsocentrals and by showing that derepressing the AS-C induces MCs there (their Fig. 5) [3982]. Skeath et al. found that when the AS-C is broken in half by an inversion, co-expression of Ac and Sc in (embryonic) PNCs ceases, and Ac and Sc become expressed in complementary subsets of clusters [3986]. The heretical implication was that ac and sc respond independently to PNC-speciﬁc enhancers, rather than controlling certain subsets of clusters via differences in their gene products [2721]. Jos´e Luis G´omez-Skarmeta et al. described the ﬁrst mutation that eliminates sc function without disabling any AS-C enhancers (sc M6 ) [1538], thereby allowing the role of sc to be assessed unambiguously for the ﬁrst time [628]. They also produced the ﬁrst enhancer map compiled via reporter gene constructs (Fig. 3.5). By these means, they showed that the “classic” roles that had been assumed

42

IMAGINAL DISCS

Macrochaetes

a % Present

100

0

mc

PSA PDC ADC 70

y

60

PNP 50

40

null point mutation

ac

ASA PS 30

sc

20

ANP APA 10

l’sc

PPA 0

PSC ASC -10

-20kb

T2

% Present

100

0

mc

PSA PDC ADC

b

70

60

PNP 50

40

ASA PS 30

20

ANP APA 10

PPA 0

PSC ASC -10

-20kb

Delete Detach y

ac

sc

l’sc

T2

% Present

100

0

c

mc

PSA PDC ADC 70 70

60

PNP 50

40

ASA PS 30

20

ANP APA 10

PPA 0

Delete Detach ac

sc

l’sc

T2

PSC ASC -10

-20kb

CHAPTER THREE. BRISTLE PATTERNS

for ac and sc were wrong. The old AS-C paradigm toppled, and a new one emerged, as recounted below. In the ensuing “gold rush,” many researchers have been picking apart each enhancer to ascertain what transacting factors inﬂuence it (e.g., see [1380]).

In 1989, Achaete and Scute were found to mark ‘‘proneural clusters’’ As discussed above, Stern deduced that each SOP arises from a group of equipotent cells, whose competence to respond to bristle-inducing signals is conferred by ac or sc. Because prepattern signals should be congruent with the ﬁnal pattern (except for cryptic singularities), the expression of ac and sc need not be conﬁned to bristles sites. In theory, ac and sc could be transcribed ubiquitously. In fact, they are not. In 1989, Romani et al. [3637] found that ac and sc RNA is restricted to nests of cells at future bristle sites. They called these nests “proneural clusters” (PNCs) [3950]. PNCs fulﬁll Stern’s 1954 prophecy. They were resolved more clearly when Ac and Sc protein distributions were seen in 1991 [912, 3982]. Each PNC has a characteristic size, shape, time of appearance, and time of disappearance [589, 912, 1538, 3637, 3982]. 1. Ac and Sc are detectable in each cluster before the SOP(s) appears. This order is consistent with the idea that Ac and Sc enable a cell to adopt the SOP state. The number of Ac- and Sc-expressing cells increases

43

2.

3.

4.

5.

6. 7.

until SOP emergence, as does the intensity of expression. Most clusters yield only one SOP. In those cases, the nest is roughly circular. Cells near the center typically manifest more Ac and Sc than those near the periphery, and one of the intensely expressing central cells becomes the SOP. Stern had imagined such a central “peak” based on his embryonic ﬁeld analogy [3613, 4096--4098]. At their zenith, most single-SOP nests contain ∼20– 30 cells. This size is roughly what was expected based on PDC shifts in ac mosaics. The PSA nest is exceptionally small (∼10 cells). The DC and SC clusters produce two SOPs each. In each case, the posterior SOP develops ﬁrst. The PDC arises at a clearly eccentric location within its nest [3373], as does the APA. Eccentric SOPs rule out certain models for how the SOP is selected. In each PNC, Ac and Sc reach higher levels in a subcluster of ∼6 cells. Among those few cells one cell attains a still greater level, which evidently boosts it above some sort of “ignition threshold” because it inevitably becomes the SOP [637, 918]. In nests that produce two SOPs, the subclusters that precede the SOPs are spatially and temporally distinct. The location of the SOP(s) within a PNC is virtually the same from ﬂy to ﬂy. Ac and Sc fade below the limits of detection before the SOP begins dividing. Presumably, the levels of Ac and Sc can wane because they have triggered the

FIGURE 3.5. Roles of achaete and scute in bristle development on the notum.

a. Phenotype of sc M6 , a sc null with a nonsense-codon mutation (“X”) before the HLH domain [1538]. Unlike previous nulls, this allele is not associated with deletions or inversions that confound gene-phenotype analyses (e.g., b, c). Bars indicate frequencies of MC presence. Sites are listed in enhancer order (dashed lines; cf. Fig. 3.4 for details and Fig. 2.6 for MC key; mc = microchaetes). Surprisingly, sc is not needed for bristles that were thought to depend on it (b), including scutellars (PSC, ASC) whence scute got its name [470]. b, c. “Classical” (but incorrect) roles of ac and sc as deduced from nulls that involve major DNA alterations. In each case, deletion of a gene (sc or ac) was achieved by crossing over between inverted chromosomes with breakpoints at different AS-C sites [2561]. With such complex “knock outs,” it is impossible to distinguish the effects of gene inactivation from the effects of disabling enhancers. Thus, the complementary sets of bristles affected by these constructs (b vs. c) give a false impression of the relative roles of sc and ac (cf. a). b. Phenotype of a sc null , In(1)sc8L sc4 R , where L and R denote left or right halves of inversions contributing to the recombinant. The deletion removes sc and the PNP enhancer, and the inversion moves proximal enhancers ASA, etc., away from achaete to the other (heterochromatic) end of the chromosome. Thus, the only enhancers still able to inﬂuence ac are mc, PSA, PDC, and ADC. Other PSA and ADC enhancers must exist in the deleted or detached domains (probably in the sc promoter region [2721]) because these bristles are often missing. c. Phenotype of an ac null , In(1)y3 P L sc8 R , which removes ac and enhancers for PSA, PDC, and ADC. Microchaetes may be only partly affected because their enhancers extend past the deletion [3688]. Although proximal enhancers PNP, etc., are disconnected from ac by the inversion, they remain linked to sc and therefore can apparently still function normally. All data are from [1538]. N.B.: The yellow (y) gene has its own set of cis-enhancers [1442, 1443], and mutations therein produce a menagerie of pigment patterns [1435, 3055] (as variegated as dog or cat breeds), suggesting that some of the same spatiotemporal trans-factors that control the AS-C may also regulate yellow [3056].

44

IMAGINAL DISCS

next step in SOP differentiation. The SOP must be able to regulate its levels of Ac and Sc separately from other cells of the nest because the SOP typically stops expressing Ac and Sc at a different time (earlier or later) from the remaining cells. PNC cells are arrested in the G2 phase of the cell cycle [4427] and are recognizable by higher levels of Cyclin A [2225]. Mitotic quiescence appears essential for proper selection of SOPs because premature entry into mitosis (prompted by artiﬁcial activation of string) disturbs SOP emergence [2225]. As for why cells must stop dividing to select a SOP, it is possible that apical signaling is hindered when dividing cells retract basally [3823], and the same logic may apply in the morphogenetic furrow where synchrony is also imposed [1126, 1903]. Quiescence cannot be under AS-C control because ac− sc− discs have G2-arrested clusters like the wild-type [4427]. Arrest is probably enforced by a parallel circuit: the same trans-acting factors that affect the AS-C may bind directly to a separate set of cis-enhancers at the string locus [2480], where they can suppress this mitosislicensing agent without needing Achaete or Scute. The discovery of these pools of nonmitotic cells answered the old question of why bristle vs. non-bristle lineages appear to diverge so early in development [996, 1372, 1742]: the split is not due to compartment-like lineage restrictions, but rather to withdrawal of pre-bristle cells from the cell cycle long before cell types are ﬁnalized. The above observations provide clues about how SOPs are selected. Before trying to ﬁt these puzzle pieces together, however, additional clues at the genetic level are examined below.

In 1995, the old AS-C paradigm toppled and a new one emerged The old AS-C paradigm was based mainly on breakpoint-associated ac LOF and sc LOF alleles [1360, 1453]. Given the complementarity of their phenotypes (Fig. 3.5), the expectation was that “ac-dependent” bristle sites would mainly use Ac, and “sc-dependent” sites would mainly use Sc. It thus came as a shock when ac and sc were found to be expressed together in PNCs [637, 3530, 3637]. To explain this correlation, a mutual activation (of ac by sc and of sc by ac) was invoked and a two-step process was imagined [2721, 3982, 4453]. In Step 1, separate enhancers would turn ac and sc on in complementary sets of clusters, with each site using one or the other gene as its founder. In Step 2, Ac would trans-activate sc and Sc would trans-activate ac in each cluster. Although parsi-

monious, this idea turned out to be wrong with regard to PNCs (although it does apply to SOPs, as explained below). In 1995, G´omez-Skarmeta et al. disproved this “Mutual Activation Model” and solved the Bristle Coding Enigma of the AS-C [1538]. To knock out sc function, they used a nonsense mutation (sc M6 ) that differs from all prior sc and ac nulls insofar as it does not disrupt any surrounding AS-C enhancers. Thus, it revealed the actual bristles controlled by sc – a subset that happens to differ dramatically from the sc-dependent subset as classically deﬁned (Fig. 3.5). The Mutual Activation Model predicts that Ac should only be found at acdependent sites in a sc null . In fact, Ac accumulates in all PNCs of sc M6 ﬂies. Additional disproof of the model came from inversions that abolished ac-sc coexpression by merely severing the connections between certain enhancers and either ac or sc [1538, 3984, 3986]. Therefore, coexpression is due to shared enhancers that activate both genes per cluster (cf. point 3 below), not to stimulation of ac by sc or vice versa. Such a possibility had been suspected [2062] but not tested rigorously. E(spl)C genes rely on the shared-enhancer strategy to a much more limited extent [871, 3075]. This “AS-CEpiphany” of 1995 has forced us to rethink how anatomy is controlled by ac and sc. The old and new conceptual frameworks are contrasted below: 1. The reason breakpoint-associated ac LOF and sc LOF alleles (vs. sc M6 ) remove certain subsets of bristles is not because they affect one or the other gene per se. (Indeed, ac and sc are functionally interchangeable [191, 918, 948, 3628], so different bristles cannot rely on qualitatively different Ac vs. Sc proteins.) Rather, allelic speciﬁcity is due to deletion or detachment of speciﬁc AS-C enhancers, each of which affects a definite body region (Fig. 3.5). For example, ac1 removes the PDC because the PDC enhancer is deleted (Fig. 3.1), not because the ac gene is defective, and the sc M6 phenotype shows that this enhancer does not need sc. 2. Categories of bristles [1802, 2561] and sensilla [948, 2519, 3637] that were formerly seen as being “ac dependent” or “sc dependent” must now be viewed as being governed by enhancers that happen to lie near ac or sc, respectively. For example, the scutellars were thought to be controlled by sc, but the sc M6 phenotype indicates that they are more dependent on ac (Fig. 3.5). 3. Temporal differences that were ascribed to sc vs. ac

CHAPTER THREE. BRISTLE PATTERNS

(e.g., sc-dependent MCs preceding ac-dependent mCs [1741], and sc-dependent chemosensory bristles preceding ac-dependent mechanosensory bristles [1802]) are probably due to “heterochronic” enhancers that are activated at different times [4584, 4844]. 4. The reason no model ever succeeded in parsimoniously explaining the allele-speciﬁc bristle subsets is that the order of enhancers along the AS-C is scrambled relative to the MC pattern [1453, 1538] (Fig. 3.4). There is no “MC homunculus” analogous to the “homeobox homunculus” [1805] that obeys the order of body segments [1113, 2508, 2672, 3494]. In this regard, the AS-C resembles the hairy locus, where enhancers for pair-rule stripes 1–7 (anterior to posterior) are scrambled: (3 & 4), 7, (6 & 2), 5, 1 [2376] (cf. similar chaos at even skipped [1323, 3717, 3992] and the lack of a “muscle homunculus” in mice [2336]). 5. The phenomenon of competence does not involve reception of an external inductive signal at the cell surface (Fig. 3.2). Rather, it entails AS-C enhancers that directly respond to intracellular position-speciﬁc, trans-acting factors. Because the enhancers (not the genes) are responsible for the competence, ac and sc are more properly called “proneural” than “competence” genes. Operationally, this distinction is supported by the ability of generalized ac or sc overexpression to cause extra MCs at abnormal sites [191, 635, 1348, 3628]. Such a result would not be expected if the genes merely enabled cells to respond to pre-existing signals at ﬁxed locations. 6. Now that the old puzzle of why speciﬁc ac LOF or sc LOF alleles remove certain bristles has been solved, the remaining question is how AS-C enhancers are controlled by spatial or temporal cues that are upstream in the genetic hierarchy [4584]. Much progress has been made in identifying these “prepattern” factors, and these ﬁndings are discussed in Chapters 5 and 6. Not all enhancers activate ac and sc equally, because only certain MCs are missing in the sc null point mutant (Fig. 3.5). This favoritism may be due to the kind of cognate lock-and-key afﬁnities between enhancers and promoters that are found at the adjacent yellow locus [1534], in the Bithorax Complex [275, 1821, 3975], and elsewhere [349, 2194, 2537, 2827, 3180, 3833]. In other words, transcription factors that bind certain enhancers (for sites that lack bristles in sc M6 ) would have greater afﬁnity for the sc promoter than for the ac promoter. Indeed, the sc and ac promoters differ appreciably from one another [918, 3171]. Such speciﬁcity would also explain why l’sc is

45

silent in SOPs, despite its closer proximity than ac or sc to some MC enhancers (Fig. 3.4) [3504]. Alternatively, these preferences could be mediated by factors that bend AS-C DNA into knotlike conﬁgurations where contacts are sterically possible only at certain points [275]. One candidate for such a role is the product of Dichaete. Dominant mutations in this gene have long been known to cause scLOF -like patterns of missing bristles [3405, 4178]. Dichaete is a Sox-domain protein that kinks DNA into an 85◦ angle [2635, 3707], so it might constrain the contacts between trans-acting factors (bound at cis-enhancers) and the basal transcription apparatus (bound at core promoters) [317, 1480, 1629, 3360], depending on where its binding sites are located within the AS-C. However, Dichaete cannot be a general partner in AS-C function because it is not normally expressed in wing discs [2977]. A more likely candidate there is Chip [3504], which ties DNA into loops [3600, 4456]. Still another suspect is the gargantuan Mediator complex [134, 412, 4386], which is known to bridge enhancers to promoters in other systems [1703, 2488, 2665]. The sc M6 phenotype suggests that some bristles need more Sc than Ac protein, but this cannot be true because sc and ac are functionally redundant [952, 2019, 2891]. Rather, the sum total of both proteins must dictate whether a bristle forms. If that total exceeds the threshold, then a bristle will arise. Otherwise, it will not.

Proneural ‘‘spots’’ shrink to SOP ‘‘dots’’ How is a SOP chosen within a PNC? The deciding factor seems to be the amount of Ac and Sc. As stated above, the intensity of AS-C proteins is uneven within each cluster, and the cell with the most Ac and Sc inevitably becomes the SOP [912, 1538]. The dose dependence of the selection mechanism is conﬁrmed by mosaics where ac+ sc+ (full-dose) cells become SOPs 2–4 times more often in a background of ac− sc− /++ (half-dose) cells than in a background of ac+ sc+ cells [912, 1794]. Why does one cell acquire more Ac and Sc than other cells in its cluster, and why does the SOP tend to arise at a certain site in each PNC? Genes that are hierarchically upstream of the AS-C (e.g., u-shaped and pannier [911, 1671]) are known to impose spatial biases [3966], but those biases can ﬁt several sorts of scenarios: 1. “Predestined SOP Model.” SOP sites could be ﬁxed by antecedent patterns of expression by genes that are hierarchically upstream of the AS-C (cf. Ch. 6). Sharp expression boundaries could pinpoint SOP sites within PNCs by their intersections, but whether

46

IMAGINAL DISCS

SOP

Proneural Cluster

Inhibition

Lateral Inhibition Model Mutual Inhibition Model

proneural cell Delta

Inhibition

proneural cell

Delta

Notch

AS-C

AS-C Delta

Delta

a

Notch

b c

there are enough boundaries to designate all sites is not known. Also unknown is whether such precision is attainable by diffusible signals [210, 902, 903, 2515, 3584], given the formidable task of addressing a one-cell target in a “forest” of tall, thin epithelial cells. 2. “Contest Model.” PNC cells might compete to become the SOP, perhaps via Dl-N “shouting matches” (Fig. 3.6; see below) [178]. Because SOPs arise within predictable subclusters, the competition must be biased [3954]. The bias could be spatial (e.g., proximity

Notch

AS-C

AS-C Notch

epidermal cell

SOP

d

to a diffusible activator or inhibitor of the AS-C) or temporal (e.g., starting to make Ac and Sc earlier). Accuracy need not be as great as in the previous model because a single victor will emerge as long as there are ≤6 contiguous contenders (i.e., the biasing process need only delimit 6-cell spots). 3. “Diffusible Activator Model.” If all PNC cells were to secrete a diffusible activator of the AS-C, then its concentration should intensify in the middle of the PNC and a cell there would be ﬁrst to cross the

CHAPTER THREE. BRISTLE PATTERNS

47

FIGURE 3.6. Hypothetical mechanisms for ensuring a single bristle per proneural cluster.

In the Lateral Inhibition Model (above), the SOP prevents neighbors from becoming SOPs [4095]. Arcs denote inhibitory signals that could be transmitted by diffusion or contact. Contact between nonadjacent cells would require extensions like the ﬁlopodia seen on moth scale cells [3051]. In the Mutual Inhibition Model (below), every proneural cell inhibits adjacent cells, and the SOP can be chosen later [1563, 3022, 3270]. Uncommitted proneural cells are shaded, the SOP is black, and cells that succumb to inhibition (and become epidermal) are white. a, b. Supposed interactions at different stages, according to the Mutual Inhibition Model. Cubes are individual cells, and inscribed circles are nuclei. a. Initially, all cells are equivalent in the extent to which they emit (Delta ligand) and receive (Notch receptor) inhibitory signals, but this balance is unstable because each cell’s AS-C feeds back on itself in a loop that traverses the other cell [1458]. The feedback is positive because the loop has two negative steps. (Arrows denote activation; cross-barred “ ” lines denote repression.) Any asymmetry will be ampliﬁed, so if one cell is slightly more inhibited, its AS-C and Delta activities will wane, as will its ability to inhibit. Ampliﬁcation is damped by a separate loop that uses the EGFR pathway (not shown) [917]. Other factors may also prevent cells from stably expressing both Notch and Delta [990, 2839]. b. Eventually this cell shuts off its AS-C and becomes epidermal. Conversely, the cell that is less inhibited keeps increasing its AS-C activity and, at some point, crosses a threshold that triggers an autocatalytic loop, which keeps the AS-C switched on. By this stage the “winner” has purged its Notch receptors, and the “loser” has purged its Delta ligands. c, d. Two of the possible outcomes for a 5-by-5 array of cells undergoing mutual inhibition [1614, 1805, 4269]. (Another is parallel stripes.) The intended outcome, a single SOP among 25 cells, is actually impossible. One way out of this dilemma is to imagine that the inhibitory arc of the loop is functional, but the excitatory arc is disabled. In that case, cells would “damp” one another but would not become SOPs, based on how severely they themselves are damped. Another means of biasing the SOP decision would then be needed, perhaps involving underlying prepattern factors or an overlay of Extramacrochaetae protein [1458]. This montage is adapted from [1458, 3866, 4115] (a, b) and [1181, 1614, 1889, 4269] (c, d). N.B.: In the wing, Delta-Notch signaling manages to create 3-cell wide stripes that become veins [989] by a feedback mechanism that seems similar [1951] but may not be (cf. Ch. 6). Thus, the same circuit may be used in different ways in different contexts [444, 627, 1457, 2018]. Odd results in the eye have raised questions, however, about the circuit’s versatility and validity [184]. See also App. 7.

“ignition threshold” to become the SOP [1462, 3584]. This scheme ﬁnds support in the intermediate level of Ac-Sc (a pre-SOP state? [912]) in subclusters where SOPs arise, but eccentric SOPs are problematic. The ﬁrst two schemes are actually components of larger hypotheses (the Lateral vs. Mutual Inhibition Models [4480]) that are discussed later. They are being treated here as stand-alone models to contrast their premises and predictions. The Predestined SOP Model is hard to reconcile with the ability of ac1 clones to displace the PDC within its cluster, because only the ordained cell should be endowed with enough Ac or Sc. Perhaps it is the highest relative (vs. absolute) level of Ac or Sc that determines the SOP. Cells within a cluster might be comparing their levels of Ac or Sc via Dl-N signaling. However, the decision cannot be purely relative because raising the absolute level of Ac or Sc can evoke extra SOPs per cluster [191, 635, 3628]. The Contest Model implies that close matches should sometimes arise in which two rivals both attain an advanced state of SOP commitment before one retreats to the default epidermal state. Indeed, pairs of SOPlike cells are occasionally seen in PNCs that ultimately

make only one bristle [1925]. How such disputes might be resolved to yield a single SOP is discussed later. The Diffusible Activator Model could explain eccentric SOPs if those nests have more of an AS-C inhibitor (e.g., Extramacrochaetae) centrally than peripherally, thus forcing the SOP to arise near the edge (like PDC displacements seen in ac1 mosaics). Some kind of diffusible factor is likely because mosaics for AS-CLOF [807, 3613, 4095, 4181] and AS-CGOF [1577, 4097] mutations often violate the rule of autonomy when the mutant/wild-type boundary skirts a bristle site. In those cases, mutant cells acquire the ability to make bristles, probably by diffusion of a rescuing agent from nearby wild-type cells [1462, 4098]. It is worth stressing that it is close neighborhood [which causes the nonautonomy] unrelated to total extent of the ac1 area which may appear as a small island just covering the bristle site or a large patch whose border happens to pass by it. This suggests a simple spread of ac+ dependent material into the ac1 tissue patch. This material then nonautonomously endows the ac1 cells with the ability to respond to the prepattern by differentiation of a bristle if only of rudimentary size. It seems to be a case of substitution therapy equivalent to the presumed support in mosaics of genedeficient cells by nondeficient tissue. [4098]

48

The rescuing molecule cannot be Ac or Sc, because they localize to the nucleus [912]. The soluble agent, which appears to be cleaved from a transmembrane precursor [3640], is probably Spitz [917] – a ligand for the EGFR pathway (cf. Ch. 6). That pathway participates in a positive feedback loop with the AS-C (see below): AS-C EGFR AS-C [917].

The SOP uses a feedback loop to raise its Ac and Sc levels Although ac-sc cross-activation was disproven for PNCs, it apparently does occur in SOPs [918, 1538, 2721]. When a 3.7-kb piece of DNA immediately upstream of the sc transcription unit is joined to a lacZ reporter, this construct is found to be expressed in SOPs but not in any other cluster cells (except at DC and PSA sites, which must have enhancers in this fragment [2721]). The same is true for a construct of the 0.8-kb ac promoter joined to lacZ. Both of these transgenes are silenced in an ac− sc− background unless ligated to PNC enhancers, implying that ac and sc genes of SOPs require Ac or Sc for autoor cross-activation. These feedback loops apparently must be primed to a threshold before they sustain themselves [917], and PNC enhancers supply the needed boost [918]. Earlier studies showed expression in all PNC cells (not just SOPs) of lacZ reporter genes driven only by the ac or sc promoter (sans PNC enhancers) [2721, 4453], but those results may have been complicated by staining artifacts or deletion of control sites that restrict expression in vivo [918, 1538]. Ac and Sc belong to the class of bHLH transcription factors that bind E boxes (Fig. 2.4). There are three E boxes in the sc promoter [918] and four in the ac promoter, although one is unresponsive to proneural proteins [3171, 4452]. The “SOP enhancer” for sc has been whittled down to a 356 b.p. piece (∼2.7 kb from the transcription start site) [918], and only one E box appears essential for auto-activation. These results imply the positive feedback loop “more {Ac and Sc} bind E boxes more {Ac and Sc},” but if this loop automatically raises Ac-Sc levels high enough to spark SOP differentiation [1538], then what prevents non-SOP cells in each PNC from doing the same? Conceivably, access of proneural proteins to E boxes in nonSOP cells is blocked by trans-acting repressors that are expressed in all PNC cells except the SOP [918, 1538]. One such candidate – Extramacrochaetae – is discussed later. Another candidate is encoded by the klumpfuss (klu) gene [2250]:

IMAGINAL DISCS

1. klu is switched on in PNCs when Ac becomes detectable but is switched off in the SOP as soon as it arises. Thus, klu’s expression could allow it to distinguish non-SOP cells from SOPs. 2. klu is genetically downstream of the AS-C (kluLOF is epistatic to acGOF in causing bristle loss), but its transcription in PNCs is independent (klu still turns on in ac− sc− ﬂies). The agents that activate klu are unknown. 3. Klu has four zinc ﬁngers and other hallmarks of a transcription factor, so it could implement SOP vs. non-SOP states directly at the DNA level. 4. Despite the above features, klu cannot be a simple selector gene because its effects appear to be nonautonomous: SOPs do arise in kluLOF mutants (as judged by high Ac levels) but fail to complete bristle development. Thus, klu causes bristle loss via an effect on a cell (SOP) that does not express klu. However, there is an autonomy-based alternative: the cell that becomes the SOP might need klu to ﬁrst turn on (leaving a legacy of other gene settings) and then off to launch the SOP on its (kluindependent) differentiative path. The AS-C must act differently in SOPs vs. non-SOPs because ubiquitous overexpression of an antagonist (hairy) quashes Ac in all proneural cells except SOPs [3982]. Further evidence for distinct (SOP vs. non-SOP) regulation is given below regarding the E(spl)-C.

Two other bHLH genes (asense and daughterless) assist SOPs A third bHLH gene in the AS-C – asense (ase) – also participates in bristle development. Like ac and sc, it is transcribed in SOPs, but unlike them it is not transcribed in other PNC cells (except on the wing edge) [438, 1079, 2039]. Its SOP expression appears late in cluster development, suggesting that it is turned on by high levels of Ac and Sc. Indeed, there are four E boxes in the 5 untranslated region of the ase transcript [918, 2039], and ase expression disappears in an ac− sc− background in all SOPs (except for the wing edge) [438, 1079]. However, ase is dispensable for most bristles, perhaps because other (AS-C?) genes compensate for its absence. In asenull ﬂies, only the wing [2039] and tergites [2690] seem affected. The stout bristles on the wing margin are often stunted, fused, or twinned, but rarely absent – implying a function for ase in bristle differentiation [1079, 2039], although the partial loss of tergital mCs suggests a role in SOP inception as well.

CHAPTER THREE. BRISTLE PATTERNS

HLH proteins typically prefer to homo- or heterodimerize with only certain partners, which in turn dictates preferences for particular DNA-binding sites [68, 348, 2089, 3017, 3179]. For example, Ac and Sc were each found to require a binding partner from outside the AS-C in order to function [588, 589, 4452, 4453]. That partner is Daughterless. As its name implies, the daughterless (da) gene was originally identiﬁed by its LOF effects on sex determination. Only later was its involvement in bristle development discovered. Flies that are doubly heterozygous for a danull allele and an AS-C deﬁciency lack some MCs, while heterozygotes for either lesion alone look wild-type [949]. This dose-dependent interaction is attributable to a need for Ac/Da and Sc/Da heterodimers, which have been shown to form in vitro [588, 589, 918, 4452]. Unlike Ac and Sc, Da is expressed ubiquitously in discs at a level that is no higher inside bristle-yielding PNCs than outside them [905, 4435]. Based on its role in the embryonic PNS, Da seems to be required for the SOP to divide [2019]. Speciﬁcally, da appears to activate cell-cycle genes in the SOP after it emerges from the cluster [949, 1755]. One downstream gene regulated by da may be Cyclin A (CycA): a CycALOF allele that causes missing MCs was induced by a P-element insertion in CycA’s ﬁrst intron, and that intron contains ﬁve E boxes [4415].

‘‘Lateral’’ or ‘‘mutual’’ inhibition ensures one SOP per PNC As mentioned earlier, Stern drew two inferences from PDC displacements in ac1 mosaics: (1) each MC SOP comes from a group of equipotent cells, and (2) the SOP inhibits its neighbors from adopting its fate [4095, 4098]. The ﬁrst conjecture was validated by the discovery of PNCs. The second was later formalized as the “Lateral Inhibition Model” (Fig. 3.6) [112, 1458, 1808, 3207, 3950]. This model made several predictions that were also supported by subsequent observations: Prediction 1:

Prediction 2:

Killing an incipient SOP should remove the inhibition and allow a different cell to become the SOP. Disabling the inhibitory mechanism should allow all the cells in a PNC to become bristles.

The ﬁrst prediction was tested in the embryonic CNS of grasshoppers, whose neuroblasts arise somewhat like SOPs of the ﬂy PNS [587, 1071, 1073, 1745, 1804]. Ablating an incipient neuroblast does indeed cause a nearby

49

cell to change its fate (from ectodermal to neural) to replace the dead one [1072, 4231]. This ability of PNCs to regenerate a progenitor cell provides direct evidence for the Lateral Inhibition Model. Consistent with the second prediction, LOF alleles of various genes cause tufts of bristles at existing MC sites (and high densities in mC zones). Most of these genes are in (or associated with) the Notch pathway [114], including Notch [1742, 1797, 3891], Delta [3277, 4859], Su(H) [3820, 3826], E(spl)-C bHLH genes [991, 1794], groucho [1794], kuzbanian [3640, 4025], shibire [1802, 1803, 3425], and big brain [1075, 3519]. Among the foregoing, big brainLOF has milder effects (even as a null), as do GOF alleles of Hairless [200, 2657] – a Su(H) antagonist. GOF alleles [191, 635, 1348, 1802] and artiﬁcial overexpression [1854, 3027, 3628] of proneural genes also cause extra bristles, but these phenotypes differ in several ways from the NotchLOF tuft syndrome [2959, 3950]: 1. Ectopic bristles arise in areas that are normally bare. 2. MCs remain separated by intervening epidermal cells. 3. Increases in mC density are less severe. The ability to put bristles into new areas (e.g., the ﬂanks of the abdomen) underscores the role of proneural genes in conferring neural competence to speciﬁc body regions [1804]. Notch-pathway genes do not share this role, with one exception. LOF alleles of groucho cause extra bristles in the wing blade and scutellum – the same areas affected by LOF alleles of hairy [1794]. The explanation is that Groucho is a promiscuous corepressor that interacts with Hairy, but not via the Notch pathway [1794, 3278]. The milder MC and mC phenotypes seen in AS-CGOF mutants [191, 1577, 1802] imply that lateral inhibition still functions, whereas it is disabled in Notch-pathway LOF mutants. Herein lies the key difference between AS-C and Notch-pathway genes. The idea that bristle cells prevent nearby cells from becoming bristles was ﬁrst proposed in 1940 by Vincent Wigglesworth [1145, 4657]. The hemipteran insect that he studied – Rhodnius prolixus – bears evenly spaced bristles on its abdomen [4663]. By comparing bristle positions in successive instars of a single individual, he found that the new bristles are added in the largest gaps of the old pattern [4657]. To explain this trend, he proposed that bristle cells deplete their vicinity of a diffusible factor that is needed for bristle initiation. This depletion prevents any new bristles from arising within a certain range of any extant bristle. Only when growth

50

pushes bristles far enough apart should an intervening cell become a bristle in the next instar. This “Local Depletion Model” is formally equivalent to the Lateral Inhibition Model, where bristle cells emit an inhibitor instead of consuming an inducer [3207, 4660]. These exclusionary (“inhibitory”) ﬁelds are thought to serve two different roles in Drosophila: 1. Fine-tuning: Inhibitory ﬁelds ensure one bristle per potential site. 2. Bristle spacing: Inhibitory ﬁelds place bristles a minimum distance apart. In more recent years, an alternative hypothesis – the “Mutual Inhibition Model” – has gained popularity [1563, 1797, 3022]. It asserts that every proneural cell (not just the SOP) inhibits its neighbors via Dl ligands. Inhibited cells reduce AS-C activity so low that they cannot become SOPs (Fig. 3.6), except for one cell (the future SOP), which somehow escapes inhibition. Like the C. elegans scheme that inspired it [3866], this model is usually depicted as a contest between equivalent cells, which adopt alternative fates after one wins and the other loses [1794, 4115]. If groups of ∼25 cells compete, then multiple SOPs should arise in checkerboard, stripe, or ring patterns (Fig. 3.6) [1889, 4269]. How can a single-SOP outcome be mandated? One way would be for the contests to occur in only a core subcluster (∼6 out of the ∼25 cells). Alternatively, the SOP could be shielded from inhibition by boosting its AS-C above a threshold for autocatalysis [443, 1458, 3022, 3823]. The two models (“LI” vs. “MI”) are debated below. Like the Contest Model discussed above, these schemes invoke cellular chitchat, but they are more concerned with preventing excess SOPs than with SOP selection per se.

Notch-pathway and proneural genes are functionally coupled The MI model envisions a feedback loop between Notch-pathway and proneural genes (Fig. 3.6). This same loop could mediate inhibition in the LI model. Some kind of cell-autonomous “N Dl” connection must exist because cells with excessive N-intra signaling in their cytoplasm lose Dl from their surface [3270]. Moreover, proneural-Notch cross-talk must exist because haplo-insufﬁciency for da enhances NLOF (split) eye phenotypes [437]. Genes wired to the end of the Notch pathway (Dl N Su(H) E(spl)-C targets?) will be repressed because E(spl)-C bHLH proteins recruit the Gro corepressor, whereas genes affected by Ac/Sc/Da should

IMAGINAL DISCS

be activated (Ac/Sc/Da targets?). If Dl were an AS-C target and the AS-C were an E(spl)-C target, then the loop would be complete. Both links appear to be functional, as the following evidence argues. However, it is important to keep in mind that showing a link under one set of circumstances does not prove that the link is used at a physiologically meaningful level in other cell types at other times [113, 871, 1330, 4584]. 1. Evidence for an “AS-C Dl” link: a. GOF: In wing discs, Dl is expressed in PNCs [2296], and ectopic expression of sc (or l’sc) activates Dl transcription in ectopic congruent areas [1854]. The transcription factor Senseless (zinc-ﬁnger class) has been implicated as a mediator of this link [3127]. b. LOF: Deleting the AS-C abolishes Dl expression in PNCs of the embryonic CNS [1672]. c. LOF: Mutating the binding sites for Ac/Da heterodimers in the Dl promoter reduces expression of a lacZ reporter gene (driven by that promoter) in PNCs of the embryonic neuroectoderm [2359]. 2. Evidence for an “E(spl)-C AS-C” link: a. GOF: In cultured cells, expression of m8 and m5 reduces the ability of Da to activate transcription of a CAT reporter gene (driven by an m8 promoter fragment that has a Da-binding site) [3171]. b. GOF: Overexpression of m8 (via a UAS-m8 transgene driven by da- or hs-Gal4) prevents SOPs from arising [3037, 4256] – an effect traceable to reduced sc transcription in SOPs (mild m8GOF ) [918] or PNCs (extreme m8GOF ) [991]. Also, bristles fail to develop when m8 is overexpressed in SOPs (UAS-m8 driven by scabrous-Gal4) [3037]. c. Correlation: E(spl)-C bHLH proteins are naturally absent from the SOP (high AS-C level) but present in the rest of the PNC (low AS-C level) [2052]. In theory, the “E(spl)-C AS-C” effect could be mediated in at least three ways [1458, 2063, 4318]: 1. Protein-protein binding via HLH domains. E(spl)-C bHLH proteins could trap Ac, Sc, or Da monomers in inert heterodimers [68, 348, 1484]. 2. Transcriptional repression of AS-C genes by DNA binding. 3. Transcriptional repression of genes that are downstream targets of AS-C proteins. The ﬁrst idea was discounted because M8 does not block Ac/Da or Sc/Da dimer formation in vitro [4452], although M8 can bind Ac, Sc, and Da under other conditions [68, 1484]. The second route seems likely for “m8 ac.” Although M8 does not bind E boxes [4452], it does bind N

CHAPTER THREE. BRISTLE PATTERNS

boxes (Fig. 2.4) [3171, 4318], and there is an N box in ac’s promoter. Indeed, m8 reduces transcription of an acpromoter CAT reporter [1794]. The “m8 sc” link should work the same way because there is an N box in sc’s “SOP enhancer” [918], but M8 still represses sc when this box is deleted. The reason seems to be that M8 can also get to the enhancer indirectly by binding an NF-κB-like factor that, in turn, binds a motif near the N box [918]. Convincing evidence for a transcriptional mode of action for M7 comes from a clever “chimera” experiment. E(spl)-C bHLH proteins are repressors mainly because the WRPW motif at their C-terminus recruits the co-repressor Groucho (Fig. 2.4) [1794]. When an artiﬁcial “m7ACT ” gene was built by replacing WRPW with a VP16 activator domain (from herpes virus [4394, 4433, 4483]) [2063] and put into ﬂies with a Gal4 driver, it caused extra bristles in various body regions – an effect opposite to the balding seen when native E(spl)-C bHLH proteins are overexpressed [918, 991, 3037, 4256]. It also caused ectopic activation of lacZ reporter genes linked to 4-kb promoter fragments from ac and sc. Apparently, m7ACT – and, by inference, native m7 – acts by inﬂuencing the AS-C at the transcriptional level (Possibility 2). The effect is apparently not exerted via E boxes because M8 does not displace AS-C/Da heterodimers from E boxes in vitro [4452]. The existence of E boxes in m8’s promoter (cf. Fig. 2.4) [171] suggests an “AS-C E(spl)-C” link [2317, 3171], and E boxes in m7 ’s promoter are required for reporter gene expression in PNCs [3974]. However, this connection is probably moot because 1. E(spl)-C bHLH genes are downregulated by N LOF , despite a functional AS-C [2052]. 2. E(spl)-C bHLH genes are activated by N GOF , even in an ac− sc− background [2052]. E(spl)-C” link were to exist, then Indeed, if an “AS-C the SOP’s abundant AS-C proteins should cause its level of E(spl)-C proteins to be the highest in the PNC, but the opposite is true: the SOP is the only PNC cell that lacks E(spl)-C bHLH proteins [2052]. The loss of these proteins from SOPs is attributable to a separate (indirect) path that is triggered by high levels of AS-C gene products: “AS-C senseless E(spl)-C” [3127]. According to the MI model, the loop between the AS-C and Notch should allow the AS-C loci of adjacent cells to interact [436, 1458, 1794]. In tracing the circuit from one cell’s AS-C into the other cell and back again (Fig. 3.6), there are two negative steps (E(spl)-C AS-C) so the overall loop is positive [1458]. For equivalent cells, the AS-C levels will be unstable. Under these conditions,

51

any stochastic change should “boomerang” (“higher AS-C loop even more AS-C” or “lower AS-C loop even less AS-C”), breaking the symmetry and tipping the balance. At some point, the victor’s AS-C level will cross a threshold, after which its autonomous SOP loop (Ac/Sc bind E boxes more Ac/Sc) will take over. Other facts are consistent with this scenario: 1. Heat-pulsing ﬂies carrying a t.s. N LOF allele upregulates sc in PNCs via an effect on sc’s “SOP enhancer” [918]. Similar stimulation of ac is seen in embryos that are defective in Notch-pathway genes [3983]. 2. In Su(H) null wing discs, ac is upregulated in PNCs [3820, 3827], supposedly via downregulation of the E(spl)-C. Interestingly, Dl is also upregulated – an effect consistent with the proposed feedback loop. Nevertheless, the circuitry inside SOPs must differ from that of ordinary PNC cells. 1. Deleting all 3 Su(H)-binding sites in the m8 promoter turns off m8 in PNCs (assayed by an m8 promoter-lacZ reporter) but turns it on in SOPs [2453]. Is the E(spl)-C regulated by genes outside the Notch pathway, and, if so, why should m8 be upregulated when deprived of activation by Su(H)? Conceivably, the deletions remove binding sites for unknown trans-acting repressors that are speciﬁc for SOPs (e.g., Senseless? [3127]). Deleting N boxes in m8’s promoter has the same effect in embryos: expression stops in PNCs and starts in neuroblasts [2317]. 2. PNCs arise normally in H null ﬂies, but SOPs fail to appear [197]. This balding is suppressed when E(spl)C bHLH genes are also deleted [197], which is strange because these genes are not normally expressed in SOPs [2052]. Possibly, H prevents Su(H) from turning E(spl)-C genes on in SOPs, but (for unknown reasons) has no effect in other PNC cells. Disabling H would cause “Su(H) E(spl)-C AS-C” (and stiﬂing of SOPs), but additionally disabling E(spl)-C would restore AS-C function [197]. These results have been interpreted as showing that the SOP is sensitive to inhibition from non-SOP cells [197] – a conclusion that would invalidate the LI model. However, the H null and double-mutant traits can also be explained by effects solely inside the SOP (given the circuitry above and the H Su(H) antagonism discussed in Ch. 2), so the facts do not inherently favor either model. (A similar restoration of normalcy is seen in the H LOF Su(H) LOF double mutant [3827].) Moreover, H transcripts are uniform in the epidermis when SOPs emerge [200], so it is unclear how H could be acting differently in SOP vs. non-SOP cells [3827].

52

3. The zinc-ﬁnger transcription factor Senseless (Sens) accumulates only in SOPs [3127]. Because (1) E boxes reside in the sens promoter and (2) binding sites for Sens (AAATCA) exist in the cis-regulatory regions of ac and sc (as well as m8), Sens could be mediating the positive feedback loop that fosters the surge of proneural proteins in SOPs: “AS-C Sens AS-C”. Indeed, sens exhibits strong synergy (i.e., dense tufts of bristles) with scute when they are co-expressed (via dpp-Gal4) in eye discs [3127]. 4. Overexpression of the EGFR inhibitor Argos suppresses SOPs without affecting PNCs [917]. It does so by blocking an autocrine loop: “AS-C EGFR ASC”. The blockage prevents the SOP from exceeding the level of Ac or Sc that is present in the PNC. The external ligand in this loop appears to be Spitz [917].

Doses of Notch-pathway genes can bias the SOP decision As stated above, SOP selection depends on AS-C dose in mosaics: a wild-type cell is 2–4 times more likely to become a SOP than a cell with half that dose (ac− sc− / ++) [912, 1794]. When the boundary between high- and low-dose cells bisects a PNC, the two genotypes presumably compete within the mosaic PNC, and their abilities to attain the SOP state are reﬂected in the overall frequencies of differently marked bristles. Such contests are a key part of the MI model, but they can also ﬁt the LI model. Maybe cells with more Ac and Sc are simply getting a head start in the race to become a SOP [917]. Such a temporal advantage might explain why female cells become SOPs 4 times more often than male cells at gender boundaries in gynandromorphs [1800]. Considering the ties between Notch-pathway genes and the AS-C, it is not surprising that doses of those genes also affect SOP choice. Notch, Delta, and Hairless are among the 21 ﬂy genes (excluding Minutes) that are haplo-insufﬁcient [118]. In haploid dose, both Notch and Delta cause more mCs [997, 4466], while a haploid dose of Hairless stiﬂes some MCs, reduces the number of mCs, and transforms cell types within the bristle organ (cf. Ch. 2) [997]. Neuroblast commitment in the embryonic CNS was already known to depend on the doses of Notchpathway genes [1000, 4467] when, in 1991, two labs reported that N dosage can bias SOP fates in discs. Garc´ıaBellido’s team induced marked +/+ clones in Nnull /+ vs. +/+ ﬂies and found more +/+ bristles in the +/+ controls (N dosage of clone:background = 2:2) than in the Nnull /+ ﬂies (2:1) [996]. In other words, the more N a cell has relative to its neighbors, the less likely it is

IMAGINAL DISCS

to become a SOP [112, 1458, 4193]. This same conclusion was reached by Heitzler and Simpson: in N mosaics, the bristles typically develop from cells having less N wherever tissues of unequal dose (1:2 or 2:3) confront one another at a genotype boundary [1797]. In Dl mosaics, the border bristles tend to come from cells with more Dl (4 vs. 3 doses) [1797], while in Su(H) mosaics, they tend to arise from cells with less Su(H) (2:4 or 3:4) [3820]. In summary, PNC cells compete to become SOPs, and the outcome can be affected by virtually any component in the ASC-Notch feedback loop. The dosage sensitivity of the SOP decision may help explain an odd class of dominant N alleles called Abruptex (N Ax ) [997, 1274, 3435, 3436]. Like artiﬁcially expressed N-intra, the N Ax mutations eliminate bristles by stiﬂing SOP initiation [197, 1798, 2627, 3235, 3863]. Unlike N-intra, however, their effects are ligand dependent: N Ax phenotypes are suppressed by DlLOF [992, 4780] or lower Dl dosage [460, 1798]. N Ax mutations localize to EGF-like repeats 24-29 in N’s extracellular domain [1747, 2182], while N-Dl binding relies on repeats 11 and 12 (Fig. 2.3) [985, 3544]. Nevertheless, for some reason, NAx proteins have less afﬁnity for Dl [2543]. Attempts to solve this “Abruptex Paradox” have invoked effects on N-N oligomerization [2182] or cis N-Dl interactions on the same cell [1798], but the only certain culprit is Fringe, an auxiliary ligand [992, 2543]. Fringe normally inhibits N’s response to Serrate (cf. Ch. 6), but N Ax mutations enable Fringe to activate the pathway [990, 994, 2096], possibly by hypersensitizing N to Dl [3245]. The ways in which N Ax traits are modiﬁable by downstream genes (e.g., H or gro [992]) or by AS-C dosage [996] suggest a GOF effect of N Ax on the pathway, as does the ability of excess N to mimic N Ax traits [3235, 3863]. However, NAx cannot mimic N’s ability to bypass shibire LOF [3863], and certain N Ax alleles must be LOFs based on their enhancement over a deﬁciency [460]. Surprisingly, the latter N Ax alleles do not act via Su(H), but rather via the Wingless pathway [461]. This cross-talk (which precedes PNCs) may involve binding between N and Dishevelled – a transducer for Wingless [151, 356, 1054]. Dl appears to be the sole ligand for N in setting SOP fates in PNCs because Dl null somatic clones exhibit a maximal tuft phenotype [4859]. Thus, Serrate is not an auxiliary ligand here, unlike its redundant role inside the 5-cell bristle organ (cf. Ch. 2).

Extra SOPs could be inhibited by contact or diffusion (or both) PNCs that produce a single MC sometimes contain 2 or 3 SOP-like cells, as detected by low-level expression of a neural lacZ reporter [1925]. The implication is

CHAPTER THREE. BRISTLE PATTERNS

that these “pre-SOPs” are competing [918, 1458]. Presumably, one eventually wins and proceeds to form the bristle, while the other loses and regresses to an epidermal state [918]. Curiously, some pre-SOP rivals are nonadjacent (Fig. 4c of [1925]). How can they compete via a Dl-N mechanism if they do not touch? This ﬁnding raises a key question. If Dl alone transmits inhibitory signals in PNCs, then inhibition should require direct contact, unless Dl can diffuse (see below). Given how epidermal cells are packed [4641], a SOP should touch no more than 5 or 6 cells [3641]. Until 1991 (when Ac and Sc were seen in situ), the number of PNC cells per MC site was thought to be in this range (based on the number of bristles per tuft in partial-LOF N and Dl mutants [3641, 3950] though shi LOF is more extreme [3425]), but then it was found that a mature PNC actually contains 20–30 cells [912, 3982] – far too many for an ordinary cell to reach [1458, 3272]. Two possible ways out of this dilemma have been proposed: 1. The only PNC cells that are truly in contention to become a SOP are those that express Ac and Sc at a higher level [918, 1458, 3022]. The number of such cells per PNC is indeed in the 5 or 6 range [912, 913, 3982]. In PNCs that yield two MCs (e.g., the dorsocentrals), the posterior SOP arises before the anterior SOP, and the PNC area that stains intensely for Ac and Sc appears to shift accordingly (Fig. 3.3). “Heterochronic pairs” [1925] are governed by single AS-C enhancers [1538]. The timing mechanism is unknown, but it too may rely on the AS-C [1925] because less Ac can cause the dorsocentral PNC to yield a single bristle that occupies an intermediate position [4095], and other odd spatial and temporal shifts affect SOPs when AS-C function is derepressed [1926]. 2. The SOP inhibits other PNC cells via ﬁlopodial extensions of its cell body [1802, 1810, 1813]. Filopodia indeed extend from SOPs before they divide [2387, 3630], and the ﬁlopodia of neighboring SOPs on the notum can touch (F. Roegiers, pers. comm.). Anti-Dl antibodies do not detect such extensions in situ [2296, 3270], but their visibility may be limited by their size. Intriguingly, when Dl- and N-expressing cells are cocultured in vitro, ﬁlopodia sprout from the former but not the latter: We consistently noted a morphological difference between Delta+ and Notch+ cells in mixed aggregates that were incubated overnight. Delta+ cells often had long extensions that completely surrounded adjacent Notch+ cells, while Notch+ cells were almost always rounded in appearance without noticeable cytoplasmic extensions. [1204]

53

In 1999, the old dogma of contact-mediated Dl-N signaling [2296, 3022] was challenged with the discovery of a soluble in vivo form of Dl that is snipped from fulllength protein by a protease [3479]. The protease is encoded by kuzbanian (kuz). Kuz may also cleave N [2298, 3153, 3239, 3640] ([3479] sequel), but N appears to be cleaved by a different metalloprotease [491, 2995], as well as by Presenilin [4156] and by a Furin-like convertase (cf. Fig. 2.3) [113, 712, 1723, 2589, 3335]. Somatic clones of kuz null cells exhibit bizarre phenotypes. The clone interior is devoid of bristles, but dense tufts of kuz null bristles develop at the borders wherever a MC or mC site is encountered [3640]. The phenotype can be explained if Kuz plays both a proneural and an inhibitory role [3640]. Conceivably, (1) a Kuz-dependent proneural factor “X” diffuses several cell diameters from the wild-type tissue and (2) X prompts kuz null SOPs to arise at the border, but (3) those SOPs cannot inhibit other PNC cells because they cannot make a Kuzdependent inhibitor (Dl). Assuming that Kuz’s main duty is to liberate Dl, it is easy to see why kuz LOF mutations have no effect on cell fates inside the bristle organ (App. 3): in that case (unlike the PNC situation), Dl only needs to activate N receptors on an adjacent cell and hence would not need Kuz to help it diffuse. Certainly, Dl is needed for inhibition, but it may not be the actual diffusible signal. Indeed, the notion of diffusible Dl is hard to reconcile with what has been learned about Dl-N interactions in wing veins and ommatidia. In those tissues, Dl cannot activate N unless cells express membrane-bound Dl, which grabs N’s extracellular domain from apposed cell membranes and escorts it into the Dl-on cell via endocytosis [3271]. The idea also seems incompatible with the antagonistic properties of artiﬁcially secreted Dl constructs [4207].

Scabrous may be the diffusible SOP inhibitor If the AS-C-Notch-pathway loop were solely responsible for ensuring one SOP per PNC, then the same phenotypes should be attainable by all genes in the loop. As mentioned before, however, LOF mutations of Notchpathway genes cause denser bristles than AS-CGOF mutations. Indeed, even when AS-C genes are overexpressed using powerful promoters, no NLOF -like tufts are ever seen [918, 3628]]. Why don’t high AS-C amplitudes drive the loop to saturation and transform every PNC cell into a SOP? Conceivably, even under these extreme conditions, the SOP still emits an inhibitor that prevents SOPs from arising closer than a few cell diameters. Indeed, this assumption was axiomatic in the LI model [3641].

54

IMAGINAL DISCS

-30 hrs

Diploid

a

b

-24 -21

A NP PS P N P

Haploid

-27

A S A

P S A

A P A

-18 Time when -15 SOP appears -12 -9 A D C P P D P C A A S C

-6 -3 0 (Pupariation) +3 hrs

P S C

Proneural Cluster

c

d

Inhibitory Field

e

f

CHAPTER THREE. BRISTLE PATTERNS

The irrepressible inhibitor may be Scabrous (Sca). LOF alleles of sca cause extra bristles near normal sites [1921, 2885], and Sca has attributes (listed below) appropriate for this role [178]. Paradoxically, however, its GOF effects in the wing are opposite to what would be expected: excess Sca inhibits Notch signaling by interfering with Dl’s activation of N [2462]. Sca does not appear to be a ligand for N [186, 2461]: when phage libraries of ﬂy DNA sequences were screened for N-binding proteins, Dl, Serrate, wingless, big brain, and N were ﬁshed out, but sca was not [4823]. Nevertheless, Sca does form complexes with N [3456], and sca interacts genetically with both Dl and N [179, 437, 1921, 2885, 3490]. 1. Sca is a secreted protein (soluble glycoprotein dimer) that travels at least 3 cell diameters from its source cells (R8 photoreceptor precursors) in the eye disc [2461]. A comparable diffusion range would sufﬁce for it to act as an inhibitor for bristle SOPs. 2. Sca is ﬁrst expressed in PNCs and strongly in SOPs, after which it fades from all PNC cells, except the SOP [197, 2885]. This time course mimics Ac and Sc. Genetically, sca is downstream of the AS-C: (1) excess Ac (due to Hairy wing mutations) causes sca expression in the same (normally bare) areas as ac (e.g., posterior wing blade) [2885], and (2) sca transcription ceases in an ac− sc− background [3974]. This “AS-C sca” link is due to 4 E boxes in sca’s promoter [3974]. As AS-C levels rise in the SOP, Ac/Sc/Da should turn on sca by binding these E boxes. The “AS-C Dl” link (described earlier) should also cause upregulation of Dl, which likewise has E boxes (≥12) in its promoter [2359]. However, incipient SOPs do not express

55

more Dl than other PNC cells [2296, 3270]. This uniformity was interpreted as favoring the MI over the LI Model [3022, 3270], but this fact is moot if a separate inhibitor is operating. According to one report [1203], N fades from SOPs as they mature, and this “deafness” should allow AS-C levels in the SOP to climb without interference from its Dl-expressing neighbors. Is the SOP similarly “deaf” to its own Sca? The answer must await identiﬁcation of the Sca receptor and determination of whether it is expressed in the SOP [2461, 3456]. One remaining mystery is why the sca null phenotype is so mild. Conceivably, additional SOP inhibitors might be acting redundantly with Sca [1158, 2885]. Sca must be as important as Dl [2592, 2628] because much of the second chromosome’s naturally occurring variation in bristle number (32% for abdominal and 21% for sternopleural bristles) maps to this locus [2172, 2381], and sca contributes heavily to the modiﬁcation of bristle number in artiﬁcial selection trials [1658, 2638, 2643, 2644].

Inhibitory fields dictate the spacing intervals of microchaetes If Sca is a SOP inhibitor, then it should suppress wouldbe SOPs within a ∼3-cell radius. In a hexagonally packed epithelium, an “inhibitory ﬁeld” of this size would cover 36 cells (6, 12, and 18 cells in concentric shells) – enough to reach all 20–30 cells in an average PNC (Fig. 3.7). Regardless of how Sca acts, it is worth pondering how inhibition might work in general. If each SOP creates its own inhibitory ﬁeld, then patterns could be constructed “dot by dot.” That is, extant SOPs could be used as points of reference for nascent SOPs in the same way that a

FIGURE 3.7. Origin of macrochaete SOPs on the notum.

a, b. Heminota (minus humeri) composed of diploid (2n, a) vs. haploid (1n, b) cells should look grossly alike because ploidy does not alter body size in animal species [1190, 1191]. The chief effect of ploidy is a proportional change in cell volume [1801]. Ovals denote macrochaete PNCs, most of which arise (asynchronously) before the notal region acquires the shape shown (cf. Fig. 2.6 for acronyms and Fig. 3.3 for a realistic depiction of a PNC). Dashed lines mark when each SOP appears [1925, 3374]. This temporal sequence is essentially irrelevant to the pattern of axonal fasciculation within the disc [4431, 4432] and axonal projection within the CNS [1455] (not shown). c, d. Area where the PDC macrochaete arises (enlarged). Ploidy uncouples mechanisms that depend on cell size (inhibitory ﬁeld diameter?) from those that do not (PNC diameter?). Assuming that PNCs (dark shading) come from a disc’s global prepattern, a 1n PNC will have the same area as a 2n PNC, but should contain more cells (small hexagons) because 1n cells are ∼21% smaller in diameter (0.793 = 0.5; 1n wing cells have 40% less area [3750]). Central cell (black with white outline) is the SOP. e, f. SOPs may use a diffusible inhibitor (circle = range) to prevent excess SOPs. If smaller cells secrete less inhibitor, then 1n SOPs may not be able to inhibit outlying cells, some of which will become SOPs and set up their own ﬁelds. Indeed, 1n-2n mosaics do have extra bristles at MC sites (and higher bristle densities overall) when occupied by 1n tissue [3750, 3754]. However, there is usually only one extra bristle per site and it tends to arise anteriorly or posteriorly, but not laterally. This result favors the Lateral Inhibition Model over the Mutual Inhibition Model (Fig. 3.6; see text).

56

IMAGINAL DISCS

geometer uses a compass for triangulation [800, 4657]. This supposition leads to two predictions: Prediction 1: Prediction 2:

The inhibitory radius should depend on the size of the SOP cell. Inhibitory ﬁelds set the intervals between evenly spaced bristles.

Both expectations ﬁt the LI model, but neither of them easily ﬁts the MI model. Prediction 1 is testable by using ploidy to uncouple cell size from body size. As a rule in animal species, alterations in ploidy cause ubiquitous (proportional) changes in cell volume but do not change body size [855, 1190, 1191]. Thus, the SOPs in haploid (1n) ﬂies should be half the volume of SOPs in diploid (2n) ﬂies, but 1n PNCs should be as large as 2n PNCs because PNC dimensions should depend only on body size. Under these conditions, a 1n SOP might not be able to secrete enough inhibitor to cover its PNC, in which case multiple bristles should arise (Fig. 3.7). Indeed, extra MCs develop at the normal MC sites when those sites are occupied by 1n tissue in 1n/2n chimeras [3750]. This result is consistent with Prediction 1. Prediction 2 is supported by the observations on the spatial distribution of mCs under various conditions: 1. In the 1n/2n chimeras mentioned above, the mCs are more closely spaced in the 1n areas [3750, 3754], presumably because 1n SOPs exude less inhibitor than 2n SOPs. 2. Triploid ﬂies have larger cells and larger bristle intervals than diploid ﬂies [1801], presumably because 3n SOPs exude more inhibitor than 2n SOPs. 3. Starved ﬂies have smaller cells and smaller bristle intervals [1801, 3617]. 4. Egfr LOF mutations cause smaller cells and denser bristles [1042]. 5. On the abdominal sternites of wild-type ﬂies, bristle lengths and spacing vary, and “nearest-neighbor distance” is proportional to bristle length [804]. Tergites exhibit a similar trend [801, 2083, 2690], and bristle interval varies with bristle length among bristle rows on the second-leg basitarsus (Fig. 5.1e) [1803]. 6. Females, which are larger than males, tend to have more bristles (except MCs) on the notum [1741, 4428], wing [1741], legs [1714], tergites [801, 1357, 2690], sternites [3522, 3555], and sternopleura [3522]. 7. Sternopleurae exhibit extra bristles (within a gender) when cell number increases [1471, 4036].

The above trends argue that mCs are spaced by a “bottom-up” process (“ ” denotes “causes”) [862]: inhibitory ﬁeld radius SOP size next bristle pattern.

distance to

In contrast, MCs appear to be patterned by a “top-down” strategy: Disc-wide coordinate system(?) pattern of SOPs.

pattern of PNCs

Here, we encounter a “MC vs. mC Paradox.” How can MCs and mCs be patterned differently if they each come from the same sort of PNC? The answer is: They don’t. In fact, mCs tend to come from proneural areas that are larger than the PNCs for MCs (see below) [1453], and the locations of the SOPs within those areas are not preset [913, 3954]. This indeterminacy explains the variability of mC numbers and positions (1) from ﬂy to ﬂy [3522], (2) on the two sides of the same ﬂy [432, 850, 1714, 3553, 3554], and (3) from one row to the next [1808, 1813, 1883]. In contrast, MCs are notoriously constant in pattern [3206, 3966]. Our insights into why MC and mC patterns behave differently gelled in the 1990s, but Curt Stern guessed the answer in 1956 from his studies of Minute LOF mosaics. As mentioned in Ch. 2, Minutes are genes whose LOF alleles cause small bristles [118, 2561]. Stern found several mosaics where a “not-Minute” (+/+) clone formed an extra MC near a “Minute” (Minute LOF /+) dorsocentral MC [4097] (cf. his Fig. 4). His interpretation resembles the argument above for 1n/2n chimeras insofar as he attributes the duplicate MCs to inadequate inhibition. In this same passage, he explains how asynchrony – if the hiatus is long enough – could permit the inhibition from one SOP to fade before a new bristle is initiated at virtually the same site (italics are author’s): All cases of duplicate or supernumerary bristles in mosaics may be interpreted by the assumptions (1) that the prepattern deﬁnes not sharp points but larger regions characterized by gradients in strength, (2) that the differentiation of a bristle tends to inhibit the differentiation of other bristles in its surroundings and (3) that under equal circumstances the inhibiting effect of a differentiating large not-Minute bristle is stronger than that of a Minute bristle. Therefore, if a Minute bristle develops in a mosaic and the more competent notMinute tissue covers nearby parts of the prepatterned region, then the inhibition exerted by the differentiating Minute bristle is insufficient to suppress the formation of a twin or supernumerary bristle in the [clone]. . . . [But] difficulties remain. Inhibition of bristles should be a mutual phenomenon. Why then, when it is assumed that a Minute bristle is not strong enough

CHAPTER THREE. BRISTLE PATTERNS

to inhibit a not-Minute twin, should not a not-Minute bristle inhibit the formation of its Minute twin? One might speculate that twin bristles arise when the Minute partner happens to begin differentiation earlier than the not-Minute duplicate and once having started on its developmental path is not subject to inhibition any more.

A similar rule governs the PNCs at MC sites where two SOPs normally develop (e.g., the ADC and PDC bristles within the dorsocentral PNC). Such SOPs are called “heterochronic pairs” because one member always arises before the other. Huang et al. argued that the interval between the two MCs is ﬁxed by the radius of the inhibitory ﬁeld that emanates from the “ﬁrst-born” SOP [1925]: In each case, the late SOP of the pair appears 3--4 cells away from the early one. The relief from inhibition of the second SOP could simply result from the intercalary growth of the disc, leading to an extension of AS-C expression beyond the radius of inhibition. . . . If our interpretation of the heterochronic pair is correct, it indicates that, in the notum and wing veins, lateral inhibition plays a dual role in pattern formation: first, ensuring that only one SOP will form at each site; second, allowing late SOPs to form at regular distances from early ones.

The “halo” of lower Ac-Sc expression around incipient SOPs (Fig. 3.3) ﬁts this scenario, because it implies an inhibitory inﬂuence from the SOP [912], although it could be an artifact of changes in cell size and nuclear position [1458]. ADC-PDC idiosyncrasies in ac1 mutants had long ago intimated an interdependence of paired SOPs: The two central bristles [ADC, PDC] together form an interdependent system. If the PDC develops, the ADC is also always present, but if the PDC is absent the ADC may or may not develop. Another property links the two central bristles together. Their positions on the thorax, as measured by the distance from some rather fixed landmarks, are variable in a peculiar way. When both bristles occur, the ADC is about where it would be in normal flies, but the PDC arises much closer to the ADC than a normal PDC would. When the PDC is absent, then the ADC does not form at its accustomed place but appears farther back. The apparent shift in position toward the posterior is clear-cut, though usually not extreme. [4096]

In summary, the PNCs at MC sites can contain more than one SOP (either naturally or under abnormal circumstances). In those cases, the interbristle distance appears to be set by the radius of an inhibitory ﬁeld that is produced by whichever member of the SOP pair arises ﬁrst. The use of an inhibitory ﬁeld in this way by a MC is unusual, but it is commonplace for mCs. The

57

histological evidence for these conclusions is reviewed below.

Microchaetes come from proneural stripes, not spots Notal mCs (and indeed most small bristles) develop after pupariation, whereas most MCs originate during the 3rd instar. For this reason, the same area can be patterned twice by separate mechanisms with no interference whatsoever [2427, 3067]. This “Heterochronic Superposition Trick” is used not only for MCs vs. mCs on the notum but also for (1) chemo- vs. mechanosensory (early vs. late) bristles on the legs [1803, 3628] and (2) photoreceptors vs. bristles (early vs. late) in the eyes [602, 603]. Between the stages of MC (early) and mC (late) patterning, the intervals between MC SOPs increase due to cell proliferation [1363, 3951] and epithelial stretching (during evagination). Therefore, the inter-SOP distances for MCs and mCs may not be as different at the outset as they appear in the adult. If so, then the radii of the inhibitory ﬁelds used by these different sets of SOPs might also be comparable. The inhibitory ﬁelds of the MC SOPs probably vanish before the patterning of mC SOPs begins [1458]. During the “classical” era of bristle-pattern research from ca 1954 (Stern’s work on ac1 ) to 1985 (Modolell’s cloning of the AS-C), most of the theorizing revolved around MCs, and this trend continued through 1991 when MC PNCs [912, 3982] and SOPs [1925] were documented. By that point, a dogma had emerged whose tenets were as follows: 1. Every bristle has its own PNC (2-SOP nests are exceptions). 2. AS-C action (conferring competence) always precedes Notch-pathway action (ensuring one SOP per PNC). In 1993 and 1997, analogous studies were made on SOPs [4428] and PNCs [3270] of notal mCs and leg bristles [3193]. These bristles, which are aligned in rows, arise differently from most MCs, although they share some features with 2-SOP MC nests. The heretical ﬁndings are listed below and summarized in Figs. 3.8 and 3.9 (cf. Fig. 3.7). 1. SOPs within a notal mC row do not originate in a ﬁxed sequence from ﬂy to ﬂy. Rather, late SOPs emerge in the largest gaps between SOPs that happen to develop ﬁrst [4428], just as Wigglesworth observed in

58

IMAGINAL DISCS

a

(N) (Dl)

54321

Dl N Dl N Dl N Dl N Dl (N)

(Dl)

(N)

(Dl)

(N)

(Dl)

(N)

(Dl)

(N)

(N) (Dl)

Dl & N Stripes appear (0-12 h AP)

b

Stripes h AP 1&5 0 3? 6-8 2 & 4? 8-12

Proneural Stripe (N) (Dl)

Inhibitory Field

Ac Ac Ac Ac Ac Dl N Dl N Dl N Dl N Dl (N)

(Dl)

(N)

(Dl)

(N)

(Dl)

(N)

(Dl)

(N)

(N) (Dl)

Ac Stripes & SOPs appear (8-12 h AP)

SOP

c

SOPs

h AP

5 1&3 2&4

8 10 12

d e

Late SOPs arise (12-20 h AP)

f

Ac

Dl

LOF

CHAPTER THREE. BRISTLE PATTERNS

hemipterans [4657]. These vagaries contradict any notion of prepatterned PNCs or predestined SOPs. 2. Unlike MCs, where Ac is expressed in PNC “spots,” notal and leg bristle rows are preceded by Ac stripes [3193, 3270]. The stripes could be chains of fused spots [3951], but other evidence argues that the distinction is real [1804, 3193, 4428]. 3. In the notal mC region, Dl and N are expressed in alternating stripes ∼8 h before Achaete is detectable [3270]. This “backward” sequence rules out an “AS-C Dl” link, and this conclusion was conﬁrmed when Dl stripes were shown to develop normally in an ac− sc− background. 4. Each Dl or N stripe is 3–5 cells wide [3270]. (Dl and N are both expressed outside their own stripes at low levels.) If Dl-N contests (Fig. 3.6) only occur between adjacent cells, then Dl and N must be using some other mechanism to form alternating stripes that are more than one cell wide. Hints of “proneural stripes” had previously come from heat-sensitive Nts1 and shits1 hypomorphs. Their notal rows 1 and 5 swell with extra bristles to stripe-like widths when pulsed at 6–12 h (Nts1 ) or 14–20 h (shits1 ) after pupariation [1742, 3863, 3891]. The NGOF allele N B suppresses all rows except row 1 [102, 2626, 2627]. Somatic clones that are null for all 7 bHLH E(spl)-C genes cause simi-

59

lar stripes of densely packed mCs [1794]. If each mC had its own PNC, then the LOF defects should have been punctate clumps (cf. Brd GOF ﬂies [2499, 2500]), rather than uniformly dense bands of bristles. The fact that Dl precedes Ac in stripes 1–5 suggests a “Dl AS-C” link (cf. proneural roles for N [460, 2540]), but the old “Dl AS-C” rule must still hold because (1) reducing Dl function during this period upregulates ac, and (2) raising N function suppresses ac [3270]. The “Dl AS-C” effect may initially be blocked within Dl stripes because Dl is seen ﬁrst on cell surfaces and only later in cytoplasmic vesicles (as is N). Derepression of ac in a Dl LOF background causes Ac to ﬁll the row 1–5 area (Fig. 3.8f) [3270]. Evidently, Ac stripes in wild-type ﬂies arise by a geometric subtraction. (The term “antineural” denotes inhibition of SOP initiation [1440, 2293, 4387, 4898].) 1 large Ac proneural area (Fig. 3.8f) minus 4 high-N antineural stripes (Fig. 3.8b) equals 5 Ac proneural stripes (Fig. 3.8c) Extra bristles in Nts1 ﬂies are mostly within the rows, which makes sense because an overall drop in N function should affect low-N (high-Dl) stripes ﬁrst, where

FIGURE 3.8. Origin of microchaete SOPs on the notum.

a. Heminotum (minus humerus), showing proneural stripes where Achaete (Ac) protein (c) foreshadows the 5 medial rows of microchaetes (e). Stripes 1 and 5 converge and fuse at their posterior tips (not shown), and other stripes develop laterally. Stripe 5 overlaps the ADC and PDC macrochaete sites, whose SOPs arise ≥12 h before pupariation (Fig. 3.7). These SOPs may still exude inhibition because no microchaetes form between them. b. Enlarged swath, showing cells (hexagons). Delta (Dl) protein appears at ∼0 h after pupariation (AP), is expressed in all 5 stripes (shaded) by 12 h AP (cf. table), and is detectable at low levels (“(Dl)”) elsewhere. Notch (N) protein is also ubiquitous but more intense (hatched) between Dl stripes. Whether these stripes arise via the kinds of AS-C cis-enhancers that govern macrochaetes is unknown [1538]. Also unknown is whether the early stripes constrain the later ones [3966]. c. By ∼8 h AP, Ac is congruent with Dl but is conﬁned to incipient SOPs (black cells outlined in white) by 12 h AP. SOPs appear in stripe 5, then 1 and 3, then 2 and 4 (cf. table). Inside each stripe, SOPs arise asynchronously. d. Most SOPs are evident by 12 h AP, but some emerge as late as 20 h AP. By that time, the earlier SOPs have not formed bristles as shown here, but have divided and appear as 4-cell tetrads. Late SOPs arise in the largest spaces between extant SOPs, as Wigglesworth found in hemipterans. He imagined that SOPs stiﬂe would-be SOPs nearby by emitting an inhibitor [4660] or absorbing an inducer [4657]. Only when a competent (Ac-expressing) cell ﬁnds itself outside inhibitory ﬁelds can it become a SOP. These rules produce a roughly even spacing of bristles in the available area. Proneural stripes will yield zigzag bristle rows if the ﬁelds are as wide (3–5 cells) as the stripes. Depending on how the signal is transmitted, inhibitory ﬁelds need not be circular. Here they are drawn as ovals (c, d). e. Heminotum, showing all bristles. f. Effect of lowering Dl function on Ac expression. When t.s. Dl LOF mutants are heat-pulsed at 5–10 h AP, Ac is expressed ubiquitously except along the midline (the same band that remains bare in N LOF and shi LOF mutants [1742, 3863]). The implication is that Dl normally represses ac between the “Dl stripes.” Apparently, the interstripes default to an ac-on state in the absence of N signaling. This diagram is based on data from [3270, 4428]. N.B.: It makes sense that bristles would be positioned late in development (after proliferation has ceased) because any pattern that is created earlier would be “messed up” by scattered mitoses [695]. This system thus obeys the general rule that proliferation and differentiation are mutually exclusive in eukaryotic cells [4882].

60

IMAGINAL DISCS

h

a

h

h

h

h

b

h Stripes appear (~0 h AP)

En

V

V

posterior

D

D

4 h Ac

3 2 Ac h Ac

V 1 8 Ac h Ac

anterior

7 6 Ac h Ac

D 5 Ac h

Proneural Stripe SOP

Ac Stripes appear (5-6 h AP)

c

V

4 3 21

V

En

V

D

D

V

V

V

d Fine-tuning occurs (12-24 h AP)

e

f

Ac

Ac

N is the limiting factor [1742, 3270]. Likewise, reducing Dl function derepresses ac in the low-Dl (high-N) stripes where Dl is the limiting factor, but there appears to be only minimal Ac intensiﬁcation in the high-Dl stripes [3270].

Ac

Ac

Ac

h

LOF

A priori, one would expect a signal-receiver system to be silenced to the same extent, whether the signal or receiver is reduced [184]. Clearly, however, the Notch pathway is not muzzled as much in high-N–low-Dl as high-Dl–low-N stripes: the former suppress ac, whereas

CHAPTER THREE. BRISTLE PATTERNS

61

FIGURE 3.9. Origin of bristle SOPs on the leg.

a. Distal half of a left second leg (mid-tibia to claws), showing stripes of Hairy protein that appear at pupariation [665, 3193]. Only 3 of the 4 stripes are visible from this posterior view. Transverse stripes are omitted. Dorsal (D) and ventral (V) midlines are marked. At this stage, leg segments are actually shorter and wider than drawn here. b. Entire 360◦ surface of an enlarged swath (ﬁlleted at D midline), showing cells (hexagons) and all 4 Hairy stripes (hatched). Black bar (below) shows the extent of Engrailed (En) expression, which characterizes the posterior compartment. c. Same swath, showing 8 stripes of Achaete (Ac) protein (shaded) that all appear at 5–6 h AP (after pupariation) [3193]. These proneural stripes (each 3–4 cells wide) foreshadow the tarsal bristle rows (e). The number of bristles per row (SOPs per stripe) decreases with distance from the V midline [1714]. Positions of SOPs (black cells outlined in white) are hypothetical. Their emergence has not been examined histologically. Lineage studies indicate that SOPs in stripe 1 (at least) originate from ≥2 columns of cells because row 1 bristles can come from both compartments [1800, 2449]. SOPs in stripe 1 are depicted posterior (6 at left) or anterior (4 at right) to the boundary (white zigzag line). Inhibitory ﬁelds around SOPs are omitted. d. Inferred movements (white arrows). Lineage studies suggest that (1) bracts and bristles in each row come from separate columns of cells and (2) bristle cells move ventrally (V arrows; stripes 2–7) or dorsally (D arrows; stripes 1 and 8) to one edge of their stripe [1800, 2449]. Adjustments of this kind (previously seen with scale precursor cells on moth wings [3051]) would explain why leg bristles are more aligned than notal bristles. Other “ﬁne-tuning” shifts may occur vertically, based on abnormal spacing in mutants with misoriented bristles [1810]. The horizontal and vertical shifts have been combined into one vector per SOP, although they probably happen at different times [1803] (Fig. 3.11). e. Adult left leg (posterior view), showing bristle rows 1–4. f. Effect of eliminating hfunction on Ac: Ac now appears in 4 broad stripes, implying that the 8 Ac stripes arise by repression of ac within Hairy stripes. This “subtractive” logic is similar to the strategy used on the notum (cf. Fig. 3.8). Dl and N protein distributions have not yet been assessed for the legs. Notch signaling might be repressing ac in the other 4 interstripes.

the latter do not. The reason could reside in (1) subtleties of Dl-N stoichiometry or (2) restricted distribution of other Notch-pathway components [3270]. For example, the m8 promoter drives reporter gene expression in 5 stripes per heminotum [2453], so ac could be off in high-N stripes because m8 can only be turned on there. However, Dl might be diffusing into the high-N stripes [3479] and activating the pathway there and only there. In that case, high-Dl cells would be “deaf” to their own signals [988], perhaps due to cis interactions between Dl and N molecules on their surfaces [2008].

Hairy paints ‘‘antineural’’ stripes on the legs LOF mutations in hairy (h) cause extra bristle rows on the notum [803] and ectopic mCs on the scutellum [1981, 2959], where Hairy is strongly expressed in wild-type ﬂies [665]. Extra bristles also arise on the head, wings, legs, and ﬂanks [1802, 2561]. The added bristles vanish when hLOF is combined with ac LOF , indicating an antagonism between h and ac. In fact, many of the excess bristles disappear even when ac LOF is heterozygous [2959, 4185]. This sensitivity implies a delicate balance of proneural (ac) and antineural (h) dosages in certain regions [665]. Curiously, the bHLH protein encoded by hairy does not bind Ac, Sc, or Da [68, 4452], so hairy cannot be inhibiting the AS-C via heterodimers. Hairy belongs to the same HLH subclass as E(spl)-C bHLH proteins. Like them, it should be acting as a repressor because its WRPW terminus recruits the Gro co-repressor (Fig. 2.4) [1242, 4865]. However, (1) hLOF and gro LOF phenotypes differ

[991], (2) hairy’s interaction with sc depends on a different (“Orange”) domain [976], and (3) still another Hairy domain binds a regulatory C-terminal Binding Protein [3377, 3430, 4865] (App. 5). Hairy must operate by DNA binding because it is converted from a repressor to an activator when its WRPW tip is replaced by a VP16 activator domain [2065]. Proof that it functions exclusively in this capacity was provided by two studies in 1994 [3179, 4451] that showed the following:

1. Hairy speciﬁcally binds a GGCACGCGAC sequence in the ac promoter (∼300 b.p. upstream of the transcription start site). This same consensus was ﬁshed out from a pool of random oligonucleotides by Hairy “bait,” so it is Hairy’s optimal binding site. At least two E(spl)-C bHLH proteins (M5 and M7) can also bind the core hexamer (CACGCG) with high afﬁnity. 2. In cultured cells, Hairy represses transcription of a transfected reporter gene linked to four of these sequences. The effect is abolished when the core binding site is mutated, so Hairy must exert its repression via this site. 3. In ac− sc− [4451] or wild-type [3179] ﬂies, an ac+ transgene (with its own promoter) is repressed by the endogenous h+ but is derepressed (as assayed by extra bristles) when h+ is replaced by hLOF . This result argues that no redundant gene (besides h) represses ac via this site (despite the ability of E(spl)-C proteins to bind there in vitro). The same phenotype is obtained when the core binding site is mutated,

62

IMAGINAL DISCS

so all hairy’s repressive effects must be mediated by this cis-silencer. On the legs, Hairy stripes play the same antineural role as Notch-pathway stripes on the notum. They appear before ac turns on (Fig. 3.9b), and their removal derepresses ac (Fig. 3.9f) [665, 3193]. Unlike the notum, however, there are only half as many antineural stripes relative to proneural stripes (4 vs. 8). How is ac repressed in the four other interstripes? The Notch pathway must be involved somehow because heat pulses to Nts1 ﬂies cause extra bristles between bristle rows as well as within them [1802, 1803]. Single Ac stripes on the notum and legs can produce more than a dozen SOPs without any obvious fragmentation of the stripes into proneural “clusters” sensu stricto. Hence, MC PNCs must be special cases of a more general regional identity that has been called the proneural “ﬁeld” [913, 1804, 2890, 3127, 3193]. Proneural ﬁelds can assume various shapes and yield various numbers of SOPs. Both sorts of variations are seen to a limited extent among PNCs themselves [912]. Field shapes include not only irregular spots (PNCs) and smooth stripes (leg zones), but also the expansive rectangles that Wigglesworth had envisioned for the tergite epidermis 60 years ago [4657]. Is it necessary to visualize the prepattern only as a series of constantly located, sharply defined peaks rising from a flattened valley floor? Could not rounded domes, ridges or even plateaus be other forms in the prepattern landscape? If so, then the inhibition . . . in the neighborhood of a developing structure may conceivably be less extensive than the particular boundary imposed by the prepattern, and more than one and perhaps many structures might appear in a single element of the prepattern. The arrangement of structures within each of these regions could then be controlled by the mechanism envisaged in Wigglesworth’s . . . model. [800].

In light of what has been learned about notal and leg bristles, the old MC-based models can now be reassessed. Just as haploid tissue makes extra bristles at MC sites (Fig. 3.7), it also makes more bristles per row [3750, 3754], so the number of SOPs per stripe is probably dictated by how many inhibitory ﬁelds can ﬁt therein. In wild-type ﬂies, the inhibitory ﬁeld would cover any PNC wherein a single SOP arises, whereas inhibitory ﬁelds would be dwarfed by proneural areas that foster multiple SOPs. Within the latter areas, any synchrony of SOP inception could lead to SOPs developing too close together. Thus, asynchrony may have evolved to guarantee a certain minimum spacing interval (by forcing later SOPs to arise outside inhibitory ﬁelds of earlier

ones). That interval would precisely equal the inhibitory radius wherever SOPs develop in linear sequence [805], but only the eye appears to use a trick of this sort in the creation of its photoreceptor array (cf. Fig. 7.7) [1804, 1808]. Apparently, the Notch pathway has the same duty inside as outside the proneural notal stripes (and in proneural ﬁelds generally) – to downregulate the AS-C. Dl is probably not an SOP inhibitor here (as conjectured for MCs) because it is not elevated in mC SOPs [3270]. Notch signaling must be low enough (in the high-Dl stripes) that the AS-C can spark SOPs, which then escape N-dependent inhibition and emit a N-independent inhibitor (Scabrous?) [3270]. This working hypothesis is an amalgam of the MI and LI models in which mutual inhibition sets the stage for lateral inhibition.

Leg bristles use extra fine-tuning tricks Leg bristles are intriguing because of their precise alignment and spacing. Moreover, they are organized into diverse subpatterns, unlike the homogeneous notal mCs. The basitarsi are especially rich in modular motifs [1714], and their development has been analyzed genetically [1799, 1802, 1807, 1808, 4346]. Their features are illustrated in Fig. 3.10. The 2nd-leg basitarsus has 8 bristle rows symmetrically disposed about the dorsal-ventral plane. The number of bristles per row varies from ﬂy to ﬂy, and adjacent rows vary independently [1813, 1883]. Bristle length and interval increase with distance from the ventral midline (cf. Fig. 5.1e) [1801, 1803], implying that dorsal SOPs have larger inhibitory ﬁelds. Higher ventral density may be adaptive for ﬁner resolution in detecting stimuli where the leg grasps objects (like the human palm). The 8-row trait has been retained in the genus over a 3-fold span of basitarsal circumference [64, 4120], implying that it is under disc-wide (vs. local) control, and this conclusion is afﬁrmed by its invariance in starved ﬂies [1801]. Superimposed upon the 8 rows are 5 chemosensory bristles, whose inter-row sites are also symmetric [2544]. A sixth bristle arises in the “vacancy” (distally between rows 2 and 3) in ∼30% of legs [1802]. The 1st- and 3rd-leg basitarsi have transverse rows of pale bristles on the front or rear face, respectively. These “t-rows” are used as brushes where the legs contact the eyes (1st leg) or wings (3rd leg) during the cleaning ritual [3376, 4225, 4462]. The 2nd and 3rd legs are not dimorphic, but the male 1st leg has a “sex comb” that females lack. Because the 2nd leg is simplest, its pattern may be most primitive (cf. Ch. 8) [1807, 1883, 4095]. If so, then evolution may have cobbled together the 1st- and 3rd-leg

CHAPTER THREE. BRISTLE PATTERNS

basitarsal patterns via the following “Basitarsal Elaboration Scenario” (Fig. 3.10), starting with the basic array of 8 longitudinal rows. Figuring out how genes were rewired to make such patterns might lead to clues about how evolution tinkers with development [1361]. The following steps are largely speculative (as is the rest of this section), and readers may safely skip to the next section: 1. Evolution inserted t-rows into 1st legs (between rows 7 and 8) and 3rd legs (between rows 1 and 2). T-rows are in the anterior lineage compartment on 1st legs, but they are in the posterior compartment on 3rd legs [4076]. First legs apparently kept row 7, but 3rd legs widened their t-rows to replace row 2 [4349]. N.B.: Homeotic mutations that transform A regions of 3rd legs to resemble A regions of 1st legs fail to affect P rows [1713, 4349], thus creating legs that have two symmetric sets of t-rows. 2. Evolution suppressed part of row 1 on 3rd legs. On the 3rd leg, row 1 seems to have lost its middle bristles at the compartment boundary. The proximal few bristles and the distalmost “orphan” bristle (all of which tend to arise anterior to the A/P line [1800]) persisted, but the orphan bristle moved across the midline to abut row 8, while the ventral campaniform sensillum made a similar move (cf. misplaced sensilla in sple1 legs [1810]; Fig. 5.12g). N.B.: In Ubx bxLOF ﬂies, the A region of the 3rd leg mimics the A region of the 2nd leg: the long-lost middle bristles reappear and the orphan bristle moves back to join a restored row 1 (cf. Fig. 3c of [2449]). 3. The sex comb originated as a modiﬁed t-row. Before this step, the 1st legs of both sexes probably had the modern female-type pattern (cf. D. ananassae [4095]), wherein the proximal ends of rows 1 and 8 converge, and the second bristle in row 8 mimics the corresponding bristle of row 1 (cf. the big second and third bristles on the 3rd leg). Various types of sex combs exist in the genus [1361, 4095]. Some are merely t-rows whose bristles are thick, blunt, dark, and curved. In others, the combs are oriented longitudinally. Cell lineage studies in D. melanogaster show that its comb begins as a distal t-row, which then turns ∼90◦ (Fig. 3.11g–i) [4344]. The following conjectures pertain mainly to how the sex comb rotates during development, but they also have implications about how it evolved: a. The comb has more bristles than any t-row, but distal rows are also broader than proximal ones,

63

b.

c.

d.

e.

so a growth gradient may exist along the segment. Indeed, the comb probably rotates via growth [3446], rather than by cell migration [1357]. The rotation may push the t-row just above it since this row often bends away dorsally. When ﬂies carrying a t.s. sex-transforming mutation are shifted to high temperature, the sequence of their comb defects reveals that bristle number is set before bristle type [281]. One t-row seems to be suppressed to make room for the rotation [2978, 2979, 3566, 3987, 4344], although two t-rows might be merging to form a single comb [4109]. Intersexual basitarsi may reveal a “missing link” transitional stage [1845]. Cell death may be involved in the rotation because this region hypertrophies (an overreaction to expanded apoptosis? [3442]) in cell-lethal mutants [3442, 3964]. The “central bristle” [1714] seems homologous to the 2nd leg’s distalmost row-8 bristle. It probably belonged to the sex comb but detached and migrated to its central site in the bare triangular area. Similar detachments of row 8’s middle bristles from the t-rows may have occurred that led to their elimination (cf. the male vs. female patterns) [4344].

If this evolutionary scenario is even partly correct, then it connotes a thorough reprogramming of cell behaviors. We know virtually nothing about the genetics of those alterations. In contrast, we are beginning to understand how SOPs rearrange within incipient bristle rows (Fig. 3.11). By heat-shocking wild-type pupae, abnormal phenotypes can be induced [1803], and the sequence of their sensitive periods suggests the following stages: 1. Lateral movements reﬁne SOP alignment. One odd phenotype in h null legs is bristle misalignment [3193]. Conceivably, SOPs on wild-type legs use Hairy/Ac boundaries as guidelines for alignment. If so, then the loss of these lines in h null legs might be causing the disorder [3193]. Additional evidence for transverse SOP movements comes from cell lineage: in certain rows, the bristle SOPs arise ventral to their prospective bract cells, while in others the bristles arise relatively dorsally (Fig. 3.9) [1800, 2449]. Because each bristle eventually lines up with its bract, the SOPs must undergo directed lateral movements within each proneural stripe (Fig. 3.11a–c). Stripes 1, 2, 4, 5, 7, and 8 lack a Hairy/Ac boundary at the edge that their

64

IMAGINAL DISCS

2nd leg D

3rd leg

P

4 3 2

V 1

A 8

7

D

D

6 5

P

4 3

V

(2)

1

A 8

7

D 6 5

100 µm

evolved?

evolved? transverse rows orphan bristle

1st leg: female D

P

4 3 2

V 1 8

1st leg: male A 7 6

D

D

5

P

4 3 2

V 1 8

A 7 6

D 5

evolved?

sex comb central bristle transverse rows

CHAPTER THREE. BRISTLE PATTERNS

SOPs supposedly seek. It is thus unclear what “homing” cues these SOPs use. 2. Proximal-distal movements reﬁne SOP spacing. Although no SOP movements have yet been witnessed in ﬂies, scale precursor cells have been shown to rearrange on moth wings [3051]. Those shifts appear to be mediated by ﬁlopodial extensions from the precursor cells, and the same ﬁlopodia may help the cells within each scale row acquire their regular spacing. Evidence for a ﬁlopodium-mediated spacing mechanism in Drosophila comes from phenotypes of mutants that have abnormally oriented bristles. In the wild-type, each row’s bristles face in the same direction. When mutant bristles face one another, they tend to have an unusually large space between them, while bristles that point away from one another tend to have an unusually small space. These correlations are explicable if SOPs are repelling one another using ﬁlopodia that are longer on one side than the other (Fig. 3.11d–f) [1810]. Such ﬁlopodia have indeed been found [2387, 3630]. Short-range adjustments of these kinds could explain why bristle rows are neater on the legs than on the notum [805, 1813, 4428]. Mutant legs that fail to elongate fully

65

have bristle rows that are as jagged as notal rows [4348], so ﬁne-tuning may require proper evagination. Notal SOPs probably differentiate in situ (i.e., wherever they arise within Ac stripes), without any subsequent corrective movements to reduce the scatter. Leg SOPs might become mobile after delaminating [1745] since they (or their descendants) could then crawl freely on the underside of the epidermis without needing to displace other cells within the crowded monolayer. Consistent with this reasoning, marked bristles are often isolated from somatic clones induced on legs [544] and tergites [1371], but not the notum [521, 1373, 1797]. Within each t-row the bristle sockets touch one another, but adjacent t-rows are separated by several cell diameters. It is therefore unlikely that t-row SOPs use isodiametric inhibitory ﬁelds [4641]. Diffusion of the inhibitor could be channeled along one axis, but it seems more plausible to imagine that the t-rows and sex comb “self-assemble” from scattered SOPs that join into tandem ﬁles by homophilic adhesion. Homophilic adhesion could also explain the tandem bristles in the medial row of the wing margin [1741]. Some sort of cellular cooperation must be occurring in both cases because sex combs and marginal rows can reform when leg or wing discs are dissociated and reaggregated [1356, 3424, 4334].

FIGURE 3.10. The four types of basitarsal bristle patterns in D. melanogaster. Each segment is drawn as if slit along its dorsal

(D) midline and spread ﬂat (P, posterior; V, ventral; A, anterior), with its proximal edge at the top. Only the 1st leg is sexually dimorphic. The 2nd-leg pattern is simplest and maybe most primitive. Its basitarsus has 8 rows of mechanosensory bristles (numbered at top), each of which bears a bract (triangle) at its base. The 5 curved “bractless” bristles reside between the rows and are chemosensory [3061]. Also shown are sensilla campaniformia (stretch detectors [4342, 4841, 4887], white circles) and trichoid hairs (small “v”s). Between rows 1 and 8 almost every cell makes a hair, so the hair density indicates cell size on the segment. There are ∼2,000 epidermal cells on the 2nd-leg basitarsus [1801]. The 1st- and 3rd-leg basitarsi augment this basic pattern with transverse rows of lighter-colored bristles. Similar rows decorate the tibias of those legs, but most of the tibial transverse-row bristles lack bracts [1714]. Transverse rows serve as brushes for cleaning the eyes (1st leg) or wings (3rd leg), and they are ideally located (A vs. P) for this role [4462]. Bristle lengths and intervals tend to increase from V to D in all patterns. In males, the most distal transverse row rotates ∼90◦ during development to form the “sex comb” [4344], which may function during courtship [1692, 4039]. Males also differ from females in having extra bractless bristles between rows 5 and 6 [1714]. In the hypothetical scheme indicated by the arrows [1883, 4096], all three leg pairs originally resembled the 2nd-leg pattern. Transverse rows then evolved in 1st and 3rd legs of both sexes, and males acquired a sex comb. The central and orphan bristles may be vestiges of these changes (see text). Alternatively, evolution might have gone the other way (i.e., simplifying an initially complex pattern). In Polycomblike LOF mutants, the 2nd leg acquires two sets of transverse rows (not shown) [1122]: one set on its A side (plus a sex comb in males) where it becomes a 1st leg and another set on its P side where it becomes a 3rd leg (see [1713]). This “homeotic schizophrenia” is a default state (cf. Table 8.1) where the Polycomb Group of “memory genes” malfunctions, thereby allowing certain homeotic genes to stay on where they should be off [1508, 1509]. Such defaults can be equivocal, however. Indeed, an opposite type of default is seen when the latter homeotic genes are disabled [4149] – i.e., 1st and 3rd legs look like 2nd legs. Drawings are for left legs from actual wild-type ﬂies but are idealized insofar as (1) segments are not perfect cylinders, (2) hairs are not so neatly arranged, and (3) color shadings in vivo make it difﬁcult to discern where transverse rows end and longitudinal rows begin, and (4) segment widths are exaggerated to avoid bristle overlaps (lengths are accurate). See also App. 7.

66

IMAGINAL DISCS

Observation 2

Inferred cell movements

1

2

D

V

1

D

2 V

1

D

V

y

b wildLOF type sple

d

c wildLOF type sple

wildLOF type sple

1

1

1

1

2

2

2

2

3

3

a

4

4

>d

3 3

d

e

4

4

f

transverse rows

g

y sex comb

h

i

CHAPTER THREE. BRISTLE PATTERNS

The chimeric composition of the reconstituted patterns proves that physically separate (differently marked) cells can align, but it is not known whether this process requires regrowth and repatterning. Hawaiian Drosophila have a tandem ﬁle of alternating “bristle-bractbristle-bract . . . ” cells in basitarsal row 8 [1801], and this peculiar row (bracted bristles are typically separated by several ordinary epidermal cells [1598]) may have evolved via heterophilic association of bristle and bract cells.

Chemosensory leg bristles are patterned like notal macrochaetes As mentioned above, the 2nd-leg basitarsus has 8 rows of mechanosensory (MS) bristles and typically 5 chemosensory (CS) bristles between the rows (Fig. 3.10). Each MS bristle bears a bract on the proximal side of its socket, whereas CS bristles are “bractless” [1714]. In many ways, basitarsal CS bristles develop like notal MCs, while the MS bristles develop like notal mCs: 1. There is an order of magnitude fewer CS than MS bristles on the 2nd-leg basitarsus (5 vs. ∼70) [1802], analogous to the disparity between notal MCs and mCs (26 vs. ∼225) [1741]. 2. Like notal MCs [3966], the number and positions of CS bristles are constant, whereas MS bristles vary in number and position from ﬂy to ﬂy (and on the left vs. right sides of individual ﬂies) [1714, 4344].

67

3. CS bristles originate before MS bristles [1803, 3628], just as MC SOPs precede mC SOPs [1925, 4428]. Thus, patterning is essentially a two-act play on both the leg and notum: in Act I a few (CS or MC) SOPs arise in a ﬁxed array, and in Act II many more (MS or mC) SOPs arise relatively randomly inside parallel proneural zones [3954]. The ability of the later kinds of patterns to form independently of (and be superimposed upon) the earlier ones is also seen on the wing margin [353, 882, 1741] and has been proposed for the integument of other insects [2425]. 4. Under certain circumstances, both CS bristles and MCs can form rows. Although CS bristles are not arranged periodically on the basitarsus, rows of such bristles are manifest on the tarsus as a whole [1811], the tibia [149, 3141, 4481], and the wing margin [1741], and other Drosophila species have strikingly regular rows of evenly spaced CS bristles on the legs [667]. Likewise, related genera have rows of evenly spaced MCs on the notum [1354], and these rows emerge (atavistically?) in D. melanogaster when proneural potential increases [1354]. Bristle intervals in CS or MC rows are larger than in adjacent MS or mC rows, presumably because later intercalary growth pushes SOPs apart after their initial interval is created by (standard size?) inhibitory ﬁelds [1354, 2959, 3950]. 5. CS bristle sites are immune to changes in cell size or organ size (and hence appear to be part of a global

FIGURE 3.11. Movements of bristle cells inferred from various studies. Bristles are denoted by circles and bracts by triangles.

Three types of hypothetical movements are shown – none of which have actually been witnessed yet. a–c. Lateral shifts that may align SOPs in longitudinal rows [1800]. a. Sections of rows 1 and 2 from a left leg (D, dorsal; V, ventral). Wavy lines indicate the kinds of boundaries seen in gynandromorphs. Male tissue is marked with yellow LOF (y). Curiously, bristles within a row tend to be more closely related to each other than to their own (physically closer) bracts. b, c. These afﬁliations are explicable if SOPs in each proneural stripe (Fig. 3.9) migrate in one direction (dorsally for row 1, ventrally for row 2) to a different column of cells where they induce neighbors to become bracts [1800, 1808]. Shaded zones in b and c denote y+ tissue as in a (they are not Ac stripes), and triangles in b mark future bracts. d–f. Proximal-distal shifts that may ﬁne-tune SOP intervals within those rows [1803, 1810, 1813]. d. LOF mutations of spiny legs (sple) cause bristles to be misoriented along with their bracts, usually a 180◦ polarity reversal (third bristle). Strangely, intervals between bristles that face one another tend to be larger (“>d”) than wild-type distance (“d”), while spaces between bristles that face away from each other tend to be smaller (“
68

positioning system of some kind), whereas MS bristle spacing and number depend on cell size and cell number, respectively [1801]. The same is true on the notum [3966], where a prepattern seems to dictate positions for MC PNCs and mC stripes but not to “micromanage” mC SOP sites within those stripes. 6. Like notal MCs, CS bristles are more affected by sc LOF than ac LOF mutations [1802], while the opposite is true for notal mCs and MS leg bristles [1360, 1453]. For these reasons, “constellation” patterns (notal MCs and CS leg bristles) and “row” patterns (notal mCs and MS leg bristles) were thought to be fundamentally different (Fig. 3.1). It therefore came as a surprise when MCs were coaxed into a mC-like pattern by gene dose manipulations (see below), implying that both types of patterns are divergent outputs of a common mechanism.

Extramacrochaetae superimposes an uneven antineural ‘‘mask’’ The extramacrochaetae gene (emc) was isolated in a screen for mutations that interact with the AS-C in a dose-dependent way [408]: LOF mutations were sought that produce extra bristles when the ratio of the tested gene to the AS-C was 1:3 or 1:4, instead of the normal 2:2. Despite the large scale of the screen, the only lesions recovered were in emc and hairy, although Notchpathway genes have similar properties [733, 997, 2690]. Thus, all genes of this kind may now have been identiﬁed. Although emc and hairy both behave as trans-repressors of the AS-C, emc LOF alleles interact mainly with sc and affect notal MCs, whereas hLOF alleles interact mainly with ac and affect notal mCs [2959, 2960]. In other body regions their effects overlap more, and their interactions with ac and sc are less restricted [1349]. Indeed, emc’s link to MCs (whence its name) is not inherent in the protein because mCs can be removed by overexpressing emc just before their SOPs arise [4453]. Like Hairy and the E(spl)-C effectors of the Notch pathway, Emc also has an HLH motif, but it lacks a basic domain for binding DNA [1156, 1388]. This feature alone sufﬁces to explain its mode of action [213, 3083] because (1) any dimers that Emc forms with Ac, Sc, and Da will have only one basic domain, but (2) two basic domains are needed to bind DNA [286]. Indeed, Emc does block DNA binding by Ac, Sc, and Da in vitro [589, 3179, 4452, 4453], and this “inert decoy” effect can been mimicked by a Da construct whose basic domain has been deleted [1754]. Hence, Emc silences the AS-C post-translationally

IMAGINAL DISCS

by sequestering Ac, Sc, and Da in inactive heterodimers. Curiously, when a basic domain from L’sc is substituted for the corresponding piece of Emc, the chimeric protein can still block E-box binding by L’sc/Da dimers [2722]. Conceivably, the inserted domain cannot dock properly with DNA because the rest of the Emc protein has evolutionarily (through disuse?) lost the ability to orient this region at a suitable angle [2722]. By adding extra doses of sc in an emc LOF background, the number of notal MCs can be increased far above the wild-type condition. Interestingly, as new MCs arise, they form rows, as if they are now obeying patterning rules that govern mCs [2959]. Indeed, the aligned MCs shift positions as new ones are added so as to maintain even spacing, so they cannot merely be sprouting from ﬁxed “cryptic singularity” sites. How can MCs acquire organizational attributes of mCs, given that MCs and mCs come from distinct proneural ﬁelds (PNC vs. stripe)? Because some MCs reside where mC rows later develop, emc LOF mutations might just be accelerating the process of stripe formation so that it intrudes into the time window for MC SOP initiation [1926]. The following facts support this hypothesis: 1. In emcLOF mutants, two rows of ectopic MCs develop along the same dorsocentral and presutural lines [1349, 1926, 2959] where mC SOPs emerge ﬁrst [4428]. Extra MCs can also be induced in these zones by overexpressing sc ubiquitously [3628]. 2. Most MC SOPs arise earlier in emcLOF mutants than in wild-type ﬂies [1926] – indicating a general acceleration of bristle development. Indeed, the plethora of “MCs” may just be an illusion resulting from a premature initiation of SOPs. If mC shaft cells start endoreplication early enough, then they could grow to the size of MCs (cf. Ch. 2), as has been inferred for mutant tergites [2690]. 3. “Excess function” (Hairy wing) mutations in the AS-C put extra MCs along the dorsocentral line [191, 635, 1577], indicating a proneural proclivity along these corridors prior to mC SOP emergence [1926]. (See [3890] for evidence of a similar tendency on the scutellum.) 4. The overall level of Sc protein increases in emcLOF mature wing discs, and the pattern of accumulation changes [912] such that the dorsocentral PNC expands to form a vague stripe along the future anteriorposterior axis. In wild-type ﬂies that stripe does not appear until ∼8 h later [3270]. These effects, and the reduction of Sc in emc GOF discs [912], are surprising, considering that Emc is supposed to affect the AS-C

CHAPTER THREE. BRISTLE PATTERNS

post-transcriptionally, but they may be indirect (see below). Heterochronic changes of this sort may have been instrumental in bristle pattern evolution, because the emc LOF array resembles MC designs of more primitive dipterans [1354, 2959]. Emc is expressed ubiquitously [1388], so one of its functions appears to be to set a threshold which AS-C activity must exceed to evoke SOPs [637, 997, 1926, 3982] – the same sort of argument proposed earlier for the Notch pathway’s role inside PNCs (the MI model). Essentially, Emc would form an inhibitory “ocean” that lets peaks (PNCs) and ridges (stripes) of a submerged prepattern landscape rise above “sea level” at certain times to make proneural territories [918]. This metaphor is emblematic of Turing-like models that appear to be irrelevant here since they invoke diffusible molecules [1727, 2806]. However, the same rules of damped autocatalysis can govern cellular automata [1033, 1437, 2651, 2694, 4699] that may be applicable. The ocean analogy is an oversimpliﬁcation because the amount of emc RNA actually varies in a landscape of its own throughout the disc epithelium. Its density tends to be low wherever Ac and Sc are expressed [913, 4453]. Although the complementarity is only approximate, it suggests a second function for Emc: to restrict the regions where proneural ﬁelds can arise by suppressing AS-C transcription [997]. Indeed, many of the extra bristles in emc LOF mutants are located in areas that are normally bare [1349], and the AS-C is derepressed at those sites [589, 3982, 4452, 4453]. In wild-type ﬂies, emc and AS-C expression are both high in some naturally nude regions, so Emc’s antineural effect there must only be post-transcriptional [589, 913]. Heterogeneous transcription of emc inside some PNCs implies a third role (aside from its global “baseline” and “carving” duties) – to conﬁne proneural potential to a subset of cells within certain PNCs [637, 913, 2062]. In such cases, the mask (of high-level emc expression) partly overlaps the PNC, rather than skirting it. The dynamics of emc and AS-C expression inside the dorsocentral PNC are revealing in this regard: After the PDC precursor has emerged, strong emc and acsc expression co-exist in the anterior part of the cluster, an area in which the ADC precursor later appears. When this SOP arises, the emc expression has been extinguished in that area. [913]

Consistent with the idea that Emc helps “ﬁne-tune” MC sites [1458], MCs are indeed positioned less accurately

69

in emc LOF mutants than in the wild-type. Particular bristles are displaced by as much as several cell diameters [1926] – as if Emc’s absence allows the area of maximal competence to broaden and shift within the PNC. Such shifts probably also occur in mosaics wherever a PNC is bisected by the boundary separating lower from higher AS-C doses [912, 1794] because the low-dose areas would be artiﬁcially mimicking the natural role envisioned for Emc [913]. Another implication from those mosaic studies is that Emc need only reduce proneural activity by a factor of two in order to perform its spatial sharpening function. Gradients of emc transcription within certain PNCs appear steep enough to span at least a 2-fold range [913]. In summary, emc affects the AS-C in time (by damping its protein accumulation) and in space (by overlaying an uneven antineural “ﬁlter” on the epidermis). This antineural emc mask (plus the prohibitive stripes painted by N and hairy) helps explain the amazing ability of bristles to still form in normal patterns, even when AS-C genes are uncoupled from all their positionspeciﬁc enhancers [1454, 1458, 1804]. 1. When ac− sc− ﬂies carrying a hs-sc [3530, 3628] or hsase [1079] transgene are heat shocked, bristles develop mainly at normal sites, despite the proneural competence that is supposedly thereby induced throughout the epidermis. 2. Bristles still arise mostly in normal areas when these same transgenes are overexpressed in a normal (ac+ sc+ ) background [438, 1079, 3628]. 3. Similar results are obtained in both genetic backgrounds (and even if the entire AS-C is deleted) when sc or l’sc is overexpressed under Gal4-UAS control [1854]. Clearly, topographic cues outside the AS-C must be constraining proneural potential. Emc provides at least some of these cues because emc LOF (in a background where ac is derepressed) decreases the positional precision of normal MCs and increases the scatter of extra bristles in general [913]. Because emc transcription is normal in ac− sc− discs [913, 4453], its peaks and valleys must be dictated by prepattern factors at a higher echelon in the chain of command. Competence is probably also gated by mitotic quiescence [637, 4427], and the dosage of da becomes limiting when AS-C monomers supersaturate the system [438, 918, 1854]. All trans-regulators of the AS-C thus far discussed (Notch pathway, hairy, emc) exert negative control in deﬁning (or reﬁning)

70

IMAGINAL DISCS

Key

4

?

?

3

?

?

?

?

?

?

?

+

~Wild phenotype Wild genotype Missing bristles

Buffered

2 1

?

?

0 1 2 3 Doses of AS-C

4

b

0

1 2 3 Doses of ac+

3

?

?

?

?

0

1 2 3 Doses of sc+

Da

Ac

4

1

c

CtBP

PLSLV WRPW

Hairy

+ + -300 b.p.

?

?

?

Gro

Emc

Sc

+

E

-200

E

-100

ac

Buffered

ac-GOF h-LOF pyd-LOF

AH 1

PA H

0 0

+1 b.p.

AHP

2

1

Wild-type

AP HP 2

3

Genotype

time window

i

4

f

Hairy

h

E

?

CACGCG

CAGCTG

N

?

2

PLSLV WRPW

g

?

orange

e d

?

0

Phenotype

a

0

4 +

Doses of emc

Doses of hairy

Extra bristles

Less AS-C activity

More AS-C activity SOPs arise

Missing bristles

Buffered

Inhib. fields Wild-type

Buffered

Extra bristles

CHAPTER THREE. BRISTLE PATTERNS

AS-C expression. Upstream factors also exist which establish spots or stripes of proneural expression by activating the AS-C, but they tend to belong to pathways that are tailored to the needs of individual discs (see below).

Dose dependency implies that HLH proteins ‘‘compute’’ bristles When the amount of a proneural or antineural protein deviates too far from its wild-type level, the resulting pattern abnormalities can be “cured” by artiﬁcially adjusting the amount of an antagonist until the deviation is “titrated” (Fig. 3.12) [408]. (Da also has dose-dependent interactions with AS-C proteins, but they are positive [949].) Examples include 1. Higher doses of emc [1348] or h [1354] erase extra bristles caused by ac GOF mutations.

71

2. Heterozygosity for AS-C deletions erases extra bristles caused by hLOF or emc LOF or both [2960]. 3. Heterozygosity for ac LOF erases extra bristles caused by hLOF [2959, 4185]. 4. Halving the AS-C dose erases extra bristles caused by half doses of N or Dl [997, 2690]. Imbalances among pro- and antineural gene products may account for the bald patches seen in interspeciﬁc Drosophila hybrids [318, 3980, 4238, 4239]. In such ﬂies, certain SOPs emerge but then abort differentiation. Hybridization may be disturbing the relative amounts of HLH agonists vs. antagonists at various sites [3980]. Those amounts would be co-adapted within species but incompatible across species [2901]. The incompatibility presumably arose via genetic drift in the parallel lineages of ﬂies after their divergence from a common ancestor [4259].

FIGURE 3.12. How HLH proteins “compute” bristles. Proneural proteins (Ac, Sc, Da) prod cells to become SOPs, whereas antineural proteins (Hairy, Emc) restrain them. a–c. Dose-dependent interactions (see key), based on data from [408, 1348, 1354, 2959, 2960]. a. Wild-type females have two AS-Cs (one on each X), and “dosage compensation” rectiﬁes the level in males to parity by a chromatin-remodeling trick (not shown) [1633, 2183, 3760] that is poorly understood [307, 308, 3249, 3997]. The phenotype resists change when AS-C dose deviates by up to a factor of two (halved to 1 or doubled to 4), and such “buffering” is also seen with hairy+ (b) or emc+ (c). However, extra bristles arise when increases in Ac or Sc combine with decreases in Hairy or Emc. d–f. HLH dimers that affect a cell’s ability to become a SOP. Shading conforms to Fig. 2.4: basic region (black), helices (hatched), loop (gray), and remainder (“noodle”). d. Ac/Da (and Sc/Da) heterodimers stimulate transcription of ac by binding “E boxes” (“class A” sites [1242, 3179]) in the ac promoter (g): CAGCTG (−262 and −195) or CAGGTG (−58) [2722, 4452, 4453]. They also activate sc [918] (not shown). These effects are blocked by Hairy and Emc, but in different ways. e. Hairy does not dimerize with Ac, Sc, or Da [68, 4452]. Its homodimers bind upstream of ac at CACGCG (a variant “N box” or “class C” site) and repress ac transcription (g) [3179, 4451], possibly via the “orange” domain [976], but more likely by recruiting Groucho (Gro, a long-range silencer that binds WRPW) [1242, 4524, 4865], C-terminal binding protein (CtBP, a short-range quencher that binds PLSLV) [2180, 3377, 3430], or other co-repressors such as dDrap (not shown) [3648]. f. Emc sequesters Ac, Sc, and Da in dimers that are inert because Emc lacks a basic (DNA-binding) domain [589, 3179, 4452, 4453]. As a result of this “broken pliers” conﬁguration, ac and sc cannot auto- or cross-activate in Emc’s presence. Emc’s effects in vivo may be greater on sc due to when (vs. where) emc is expressed. This “disabled dimer” trick is also used in the LIM-HD and POU-HD families, where Beadex [2854, 3908] and I-POU [4380] play roles analogous to that of Emc. This process whereby an inhibitor sequesters an activator is called “squelching” [1984, 4806] to distinguish it from “quenching” (i.e., the blocking of an activator by an inhibitor after both have bound a DNA site) [1600, 1984, 2350]. g. Promoter region upstream of the ac transcription start site (crooked arrow). E and N boxes are marked; “+” = activation; “−” = repression. h. An example of buffering, where three mutations were studied singly or combined: ac GOF (A, a.k.a. Hairy wing), hairy LOF (H), and pyd LOF (P). All of them increase activity of the AS-C [635, 733, 3179, 4451]. The number of mutations per ﬂy is along the x axis; the number of extra scutellar macrochaetes is along the y axis [3069]. Phenotypes remain nearly wild-type until the double-mutant threshold (cf. a–c), whereupon synergism ensues. i. Time Window Model for pattern buffering. In each panel, the x axis is time, the upper triangle denotes gradual production of SOPs in a proneural ﬁeld, and the lower triangles denote the time needed for SOPs to extend inhibitory ﬁelds (they are not really sequential). In wild-type ﬂies (middle rectangle), the time available for patterning (window width) exceeds the time needed (upper triangle width), so slight changes in AS-C activity (2nd and 4th rectangles) may merely shift the timing of events. In contrast, more extreme changes (1st and 5th rectangles) may additionally alter the rate of accumulation of Ac and Sc (gradient slope). High-AS-C ﬂies should make SOPs so quickly that they cannot fully extend inhibitory ﬁelds before new SOPs arise nearby (leading to extra bristles), whereas low-AS-C ﬂies will make SOPs so slowly that time expires before the wild-type number is attained. Whether HLH proteins also help terminate this time window is unknown, but a HLH Hourglass has been found in vertebrates [2293, 2294] that uses proneural and antineural bHLH antagonists. See also App. 7.

72

Evolutionary drift in “volume settings” of interacting HLH genes might also explain the intersexuality of many hybrids in this genus [1174, 1882, 1883] and other insect groups [1520, 1524, 3398]. Gender-speciﬁc traits in Drosophila appear to be decided by the ratio of X chromosomes to autosomes [1173, 4077] , as measured by relative doses of gene products [3266]. Remarkably, the numerator and denominator elements include the proneural bHLH proteins Scute [1028] and Daughterless [3268] , the panneural bHLH protein Deadpan [209], and the antineural HLH protein Extramacrochaetae [1155] , suggesting that evolution may have commandeered a cassette of HLH neural-patterning genes for secondary duty in sex determination [1173, 2018, 4824]. HLH genes decide cell fates not only in PNS neurogenesis and sex determination, but also in photoreceptor initiation [182, 1076] , CNS neurogenesis [485, 3981] , myogenesis [2177] , and gliogenesis [1464] , as well as in the development of midline mesectoderm [899] , Malpighian tubule tip cells [4672] , appendage tips [1166] , tracheal tubes [4548] , salivary ducts [899] , endodermal tissues [4269], etc. [914, 1343, 1744]. Like Numb, therefore, HLH transducers function at an abstract level that transcends histotype (cf. Ch. 2). Collectively, they act like analog-to-digital transducers to ensure all-or-none outcomes [816, 2958, 3186, 4584]. In short, cells use HLH proteins to do arithmetic. The virtuosity of HLH input/output devices is illustrated by the sex-determining genes, which reliably produce discretely male or female (vs. intermediate) bristles in sex combs despite X:A ratios that range between 0.5 (male) and 1.0 (female) [4370] . The choice of bristle type is made on a cell-by-cell basis [3734, 4105] a few hours after the number of sex comb SOPs is set [281, 2106]. What kind of math is being used to convert quantitative signals into qualitative states? The “SOP Computer” seems to operate by combining positive and negative scalar inputs (Ac, Sc, Da, Emc) as HLH heterodimers. For the scheme outlined below to work, Ac and Sc must be rate limiting for dimer formation with Da and Emc. If Da/Da dimers form too readily, then Da’s presence before Ac and Sc appear [905, 4435] could be problematic (although dimer preferences could be imposed biochemically or sterically by thirdparty agents [245, 286, 2046, 3083]). Indeed, Da dimerizes only weakly with itself [588, 4453]. 1. Upstream trans-activators of ac and sc intensify at certain epidermal sites, bind site-speciﬁc enhancers near ac and sc, and start to evoke Ac and Sc at time

IMAGINAL DISCS

2.

3.

4.

5.

“t1 .” Ac and Sc form heterodimers with Da or Emc [68, 589, 4452], instead of binding each other or homodimerizing [588, 918, 1484, 3974, 4453]. In these proneural areas, the number of Ac/Da and Sc/Da heterodimers per cell will increase at a rate “r1 ” that is damped by diversion of Ac, Sc, and Da monomers into inert heterodimers with Emc [4453]. Indeed, all these heterodimers form readily [68, 589, 4452], and Emc has been shown to inhibit E-box binding [4452] and transcriptional stimulation [589, 4453] by Ac-Sc-Da combinations in a competitive, dosedependent manner. Eventually, Emc is depleted (unless Emc is resupplied via transcription), and this event marks the ﬁrst titration threshold “T1 ” [1387, 4584]. Thereafter, the rate of Ac/Da and Sc/Da accumulation will increase to “r2 ” (assuming there was initially more Da than Emc). At some point, the concentration of Ac/Da and Sc/Da heterodimers crosses a second threshold “T2 ” that triggers autocatalytic production of Ac and Sc [918] via (1) direct feedback of the heterodimers on E boxes near ac and sc [2722, 4452], and (2) indirect feedback via sens, whose promoter has E boxes and whose product binds near sc and ac [3127]. Levels of Ac and Sc rise exponentially to saturation (limited by Da) [997]. The cell becomes a SOP (via target genes whose E boxes bind Ac/Da or Sc/Da) and emits a signal that prevents nearby cells from making more Ac and Sc. SOPs continue to arise within the proneural area (in either a stochastic or a spatially biased manner) until time “t2 ,” when all competent cells have either become SOPs or are inhibited by nearby SOPs.

This algorithm does not explain why Sc transcription rises in the wing disc when Emc is reduced [912], because Emc only acts post-transcriptionally and autoregulation of AS-C loci is only supposed to occur in SOPs. Conceivably, some trans-activators of the AS-C may also be bHLH proteins that could be disabled by Emc [1387]. The only gene so far identiﬁed upstream of emc itself is polychaetoid (pyd) [4237]. Pyd’s presence in intercellular junctions suggests that emc might be activated by intercellular signaling. The HLH scheme shows how easy it is to do math with molecules by titrating monomers. It uses addition (Ac + Sc = Total proneural), subtraction (Total proneural − Emc = Net proneural), and multiplication (feedback in Step 4), but not division. Indeed, it now appears that even the famous X:A “ratio” is an illusion of a

CHAPTER THREE. BRISTLE PATTERNS

procedure that strictly relies on counting [816]. The dose sensitivities of Notch-pathway components are probably also due to dimer interactions – the most obvious of which is the docking of Dl and N in trans [1204, 3544] and possibly also in cis [2008] at cell surfaces. Titration-based arithmetic may also operate at the RNA level. A motif exists in the 3 -UTRs (untranslated regions) of transcripts from hairy, emc, the E(spl)-C genes m3, m4, m5, and mγ , and the m4-related genes Bearded, Bob (Brother of Brd = actually 3 genes A, B, and C), and Tom (Twin of m4) – all of which act antineurally [2382, 2386, 2499]. This “GY” box reads GUCUUCC (except for emc’s GUUUUCC). The complementary CAGAAGG appears in 3 -UTRs of transcripts from the proneural genes ac, l’sc, and atonal [2386]. Hence, mRNAs from these proand antineural genes could form heteroduplexes that could conceivably modulate transcription, mRNA localization, processing, turnover, or translation [233, 1219, 1702, 2686, 2920, 2965, 3400]. Docking between hairy and ac mRNAs might explain why interactions between these genes are so dose sensitive (Fig. 3.12), despite the failure of their proteins to dimerize [68, 4452]. Other motifs in these antineural-class UTRs include “Brd” (AGCUUUA) and “K” (UGUGAU) boxes, both of which decrease accumulation of transcripts [92, 2382, 2384, 2385].

Robustness of patterning may be due to a tolerant time window For Threshold T1 , what presumably matters are the absolute amount of Emc and the rate of production of Ac and Sc, and the latter rate should depend on the amount of stimulatory “prepattern” factors [2891]. Because the variables are all constitutive (with no feedback until Step 4), the stoichiometry easily explains why the process is sensitive to gene dosage. What is not so easy to understand is why phenotypes remain wild-type when the AS-C dose is halved or doubled, or when the dose of emc is halved (Fig. 3.12). Why don’t such changes affect bristle number like greater alterations do (e.g., halving emc dose plus doubling sc dose)? Historically, the ability of animals to maintain phenotypic constancy in the face of perturbations (genetic or environmental) led to the related concepts of “buffering” [1748, 3709] , “canalization” [1470] , and “robustness” [2565, 3032, 4646] . Certainly, ﬂies do use a variety of error-correction strategies at every echelon (e.g., genes [4584] , proteins [2760] , cells [1292] , and tissues [12, 1940, 2527, 3042]), and many of these “quality control” tricks are undoubtedly sophisticated. However, the buffering of SOP number may

73

instead have a trivial explanation. To wit, there may be enough tolerance in the system to accommodate minor changes in AS-C activity by letting events shift along a time line (Fig. 3.12). If the slack is ﬁnite, then greater changes in AS-C activity will eventually push events to the edges of this time window, and no further slippage will be possible. At that point, the phenotype should begin to overtly reﬂect added deviations in genotype. Speciﬁcally, 1. Two-fold increases in AS-C dose (or halving the emc dose) might merely hasten Step 5 (before t2 ). Further ﬂooding of the system with Ac and Sc (i.e., changes greater than a factor of two) might accelerate the process so much that more cells can cross T2 and become SOPs before SOP inhibitory ﬁelds can extend to their maximal diameter – hence leading to greater bristle density. 2. Halving the AS-C dose should postpone the time at which Step 5 is reached. However, if upstream transactivators (Step 1) are still present, then the process might continue beyond t2 until it reaches completion. Further decreases might slow the process so much that the prepattern factors disappear before all the gaps between SOPs are ﬁlled in. This “Time Window Model” is buttressed by timing shifts of emc LOF SOPs [912, 1926, 2690]. It is also consistent with temporal changes in (1) N null SOPs [3689], (2) the CNS and PNS of N null embryos [379, 1563], and (3) SOPs in Egfr LOF mutants [917]. On the wing, SOPs of ectopic sensilla (caused by hLOF or ac GOF ) can arise within a broad time span and still develop normally [362], and the same is true for extra SOPs in PNCs of pyd LOF mutants [733]. Like late passengers missing a train that always leaves the station on schedule, these delayed pyd LOF SOPs make shorter bristles [3067] because cuticle secretion starts on time, and a similar constraint explains the fewer bristles that come from depleted histoblast nests [581]. Indeed, AS-C and E(spl)-C genes may be short and intronless so that their proteins can be made quickly and timed precisely [92]. Figure 3.13 summarizes the core circuit for bristle SOP selection. The idea that SOP inception is time sensitive may explain why it is temperature sensitive in hypomorphs [767, 769] not only during the PNC-SOP window [768], but also much earlier [770, 1997, 1998, 3405]. As Goldschmidt argued long ago [1520], shifts in the relative rates of reactions can uncouple key processes, leading to a later failure of the system to reach a critical threshold. The

74

IMAGINAL DISCS

a

Stages

(Prepattern) Ac Sc

and Hairy Emc

or

or 1. Competence

and

Da

Non-SOP

SOP

(Proneural) or

1 1 0

0

1

c

1 or

1

0

d

1 or

0

1

1 1

1

1 1

1

2. Fine-tuning

0

0

1

Non-SOP

Ac h

or

or

and 0

1 and 1

1 1

SOP

1 1

or

1 and 1 Non-SOP 1

and 1

or

1

1 or

1 1

1

1 and

0 1

1

1 or

0 0

1

and

0 0

0

1 1

0

1 0 0

SOP

1

and 0 0

SOP

1 and

or

or Dl Sca

b

SOP

Ac Sc

and Dl Sca

0

0 0

or

antecedent reactions for bristle patterning remain to be identiﬁed. A time window also constrains the number of SOPs that form the cluster of chordotonal receptors in the femur. In contrast to how bristle SOPs arise, chordotonal SOPs emerge from their PNC as a tandem chain via reiterative induction of new SOPs by old SOPs. The process starts and stops at deﬁnite times, and the number of SOPs can be increased or decreased by raising or

e

0

lowering the perceived volume of the inductive (EGFR pathway) or inhibitory (Notch pathway) signals during that window [4898].

Atonal and Amos are proneural agents for other types of sensilla Chordotonal PNCs differ from bristle PNCs insofar as they rely on the bHLH gene atonal instead of ac or sc [3181, 4897], and the same is true for photoreceptors in

CHAPTER THREE. BRISTLE PATTERNS

75

FIGURE 3.13. Circuitry that assigns SOP (bristle) vs. non-SOP (smooth cuticle) cell fates in the epidermis (cf. Fig. 2.7 for key to

symbols). a. Core pathway of proteins that decide SOP vs. non-SOP identities (see text for further details). Abbreviations: Ac (Achaete), Da (Daughterless), Dl (Delta), Emc (Extramacrochaetae), h (hairy), Sc (Scute), Sca (Scabrous). In the ﬁrst stage, “prepattern” factors (cf. Chs. 5 and 6) prompt the synthesis of Ac and Sc in certain regions by acting through cis-enhancers in the AS-C. When Ac and Sc (which act redundantly) heterodimerize with Da (which is ubiquitous), they endow a cell with the “competence” to become a SOP. This “proneural” state can be prevented by the antineural agents Hairy (which primarily affects Ac) and Emc, both of which act intracellularly (Fig. 3.12). In the second stage (shaded), one or more SOPs are selected by “ﬁne-tuning” within the proneural ﬁeld. Continued accumulation of Ac and Sc prods a cell to become a SOP, unless it is blocked by external antineural signals emitted by SOPs themselves. Dl and Sca are thought to be short- and long-range agents, respectively (see text), that create an “inhibitory ﬁeld” around each SOP. b–e. Schematics of cells (b–d) that are selecting various fates at different tissue locations (e). States of components (cf. a) are recorded as “1” (present) or “0” (absent), and black circles denote determining factors. b. The circuit as it functions in areas where Hairy is expressed (e.g., leg stripes, e). The cell never becomes competent because Hairy is the controlling factor. c. If a cell is impelled by prepattern factors (circled “1”) and no antineural agents are present, then it will complete the gauntlet and become a SOP. d. Any cell inside the inhibitory ﬁeld of a SOP cannot become a SOP. Sca (circled “1”) is here assumed to be the diffusible inhibitor. e. Rectangular piece of leg skin (Fig. 3.9c) containing one Ac stripe (shaded) and the adjoining Hairy interstripe (hatched). Among these ∼150 cells (hexagons), only 4 SOPs (black cells, c) will form. Remaining cells are prevented from becoming SOPs because they contain Hairy (b) or are inhibited by a nearby SOP that adopted this state (stochastically) before they could do so (d). Ovals mark limits of inhibitory ﬁelds, which are assumed to be anisotropic because the inhibitor (Sca?) does not diffuse freely. N.B.: In D. melanogaster, non-SOP cells look alike, regardless of whether they were previously proneural, but other species convert some proneural ﬁelds into pigment stripes [615]. Although the circuitry depicted here is digital, the “SOP computer” actually operates partly in an analog (threshold) mode (Fig. 3.12; see text). Also, it is not known whether Sca acts via the same pathway as Dl [113], and other questions remain about the core logic [178]. In principle, the “and” gate preceding Ac and Sc should cancel out the subsequent “or” gate, but this part of the circuit is included because Ac and Sc can be independently changed by mutations (Fig. 3.5) or various LOF-GOF manipulations (App. 5).

the eye disc [182, 1076, 2041, 2042]. Still another bHLH gene – amos – establishes PNCs for olfactory sensilla and leg claws [1587]. LOF and GOF studies have shown that these different bHLH genes not only set up PNCs, but also determine the identities of the sensilla that arise therein (cf. Ch. 2). For example, overexpressing Scute induces bristles only [912, 918, 2038], whereas overexpressing Atonal induces chordotonal organs [764, 2038, 2040], and overexpressing Amos elicits both chordotonal and olfactory sensilla [1587, 1928]. Thus, no clear distinction exists between proneural genes on the one hand and sensillumidentity genes on the other [448, 1928].

Other (upstream) pathways govern bristle patterning Polychaetoid is intriguing because it may link the HLH circuitry to the Notch pathway [4811] and adherens junctions [533]. The involvement of pyd in HLH affairs is indicated by (1) reduced transcription of emc in pyd LOF wing discs [4237], and (2) synergy of pyd LOF with emc LOF , hLOF , and ac GOF [733, 1802, 3068, 3069]. The dose-dependent synergy of pyd LOF with both NLOF and DlLOF is equally dramatic [733], and other clues also point to an afﬁliation

of pyd with the Notch pathway – viz., (1) Pyd’s location at junctions [4236] in the same apicolateral ring where N is concentrated (Pyd is homologous to mammalian Zonula Occludens-1) [1203], and (2) Pyd’s ability to bind the product of a gene (canoe) that also interacts with N (App. 5) [4236]. If pyd serves such a key role, then why don’t pyd LOF mutations affect all bristles equally? Why, for example, are extra MCs found in the dorsocentral area ∼10 times more often than in the notopleural area [733]? The spatial heterogeneity cannot simply be due to LOF allele quirks (a bugbear of the AS-C) [533], because it also occurs with null alleles [733]. This same question was raised for Hairless in Ch. 2, where it was argued that a second gene might share its function. Although a redundant agent might also solve the pyd problem, it is possible that pyd is part of a separate pathway that is upstream of the entire SOP siting program [2363]. Pyd may instigate certain PNCs (via “Pyd AS-C”), but genes in other pathways may establish the other ones. Such genes are the elusive prepattern factors that Stern tried so hard to ﬁnd. How they act is discussed in subsequent chapters.

CHAPTER FOUR

Origin and Growth of Discs

Insect imaginal discs are barely visible to the naked eye, so detailed observations had to await the invention of adequate magnifying lenses [1266, 3064]. Discs were ﬁrst described by the great naturalist Jan Swammerdam (1637– 1680) [821, 3422, 4586], a contemporary of Leeuwenhoek’s, who applied his training in human anatomy to the study of insect morphology [3133]. In his Book of Nature (printed in English in 1758), Swammerdam waxes lyrical about the metamorphosis (which he calls “mutation”) of appendages (“horns” are antennae) in hymenopteran larvae (“worms” of bees):

The wings, horns, and other parts which worms without legs seem to acquire about their chests at the time of their mutation are not truly produced during the period of mutation, or, to speak more agreeably to truth, during the time of the limbs shooting or budding out, but . . . have grown there by degrees under the skin, and as the worm itself has grown by a kind of accretion of parts, and will make their appearance in it upon breaking the skin on its head or its back, and thereby give it the figure of a nymph, which it would afterwards of itself assume. Hence it is, that we can with little trouble produce [by dissection] the legs, wings, horns, and other parts of an insect, which lie hid under its skin while in the shape of a naked worm, which has neither legs nor any other limbs. . . . The nymph . . . is nothing more than a little worm, which, the growth of legs, wings, and other limbs hid under its skin being perfected by time, at last bursts that skin, and casting it off, gives us a clear and distinct view of all those parts. This change . . . is no more mysterious or surprising than what happens when one of the meanest plants, despised and trodden under foot, gradually swells on every side, and after producing a bud, by bursting the little case containing it, presents an elegant and beautiful flower. [4218]

76

The ﬁrst account of discs in dipterans appeared in 1864 when August Weismann (renowned for his germ plasm theory [1487]) analyzed the stages of development in Musca and Sarcophaga [4586]. In 1936, after D. melanogaster had become the darling of genetics, a former student of Spemann’s, Charlotte Auerbach [2215, 4696], traced the development of leg, wing, and haltere discs back to the newly hatched larva [142]. Dietrich Bodenstein studied the development of all the discs for Demerec’s Biology of Drosophila (1950) [1019], which served as the standard reference for a generation of ﬂy researchers. The precursors of discs within the embryonic blastoderm were mapped in the 1970s by microcautery [425], microbeam irradiation [2591], and cell-lineage techniques [1695, 2026, 4652]. Each disc was found to come from a distinct cluster of blastoderm cells (Fig. 4.1).

Segmentation genes set the stage for disc initiation Not until the genetic basis of segmentation began to be analyzed ca 1980, however, was it possible to discern the factors that cause discs to originate. In that year, ¨ Christiane Nusslein-Volhard and Eric Wieschaus deﬁned the hierarchy of “segmentation genes” that divides the embryo’s anterior-posterior (A-P) axis into metameres [2178, 3151, 4069]. The hierarchy has four echelons (Fig. 4.2) [2717, 3248, 3843, 4671]: 1. Axis genes (a.k.a. “maternal-effect” genes). Three proteins form gradients along the A-P axis: “Bicoid” is made from mRNA maternally deposited at the anterior pole [1104] , “Nanos” comes from posterior

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

A7

A1

T3

T2

77

T1

3

H

1

8

4

2

10

9

Discs

as T1 T2 T3 2 T 3 A1 1T T dEpi dEpl

blastoderm fate map

A7

H1

pc pmg

H2

vNR amg

H3 H4

D A

H5 H 6

mes

P

1 2 3 4 5 6 7 8 9 10

Labial Clypeolabral Humeral Eye-antennal 1st Leg 2nd Leg 3rd Leg Wing Haltere Genital

14 cells 100 µm

V

5

6

7

FIGURE 4.1. Origin of discs as “islands” within the embryonic ectoderm. Left side of an embryo (below) at the cellular blastoderm stage (D, dorsal; V, ventral; A, anterior; P, posterior). Spots indicate where discs (key at right) or histoblast nests later form. In this fate map, segments (straight or warped rectangles) are numbered in each region (Head, Thorax, Abdomen). Areas shaded medium (dEpi, dorsal epidermis) or dark (vNR, ventral neurogenic region) form larval skin, except that ∼25% of vNR cells ingress as neuroblasts. Gastrulation internalizes the midgut (amg, pmg), mesoderm (mes), pole cells (pc), and virtually the entire head [4421]. Amnioserosa (as) is extraembryonic. This fate map is based on Hartenstein’s atlas [1739], with the following exceptions: (1) D midline is not tilted toward viewer; (2) clypeolabral, eye, labial, and humeral sites are as per gynandromorph maps [1695, 1777, 2026, 4146], although the actual humeral site may be more ventral because its spiracle and the 1st-leg disc can share blastoderm cell ancestors [237, 928, 2819, 4565]; (3) eye anlage is shown as one oval instead of the three spots that Volker Hartenstein used (pers. comm.) as an arbitrary way to connote the eye disc’s mysteriously diffuse origin [1147, 1224, 2103] and its patchy apoptosis [3058, 4825]; (4) thoracic discs are more dorsal as per cell transplant data [2820]; (5) wing disc and 2nd-leg disc are fused as per histologic [827, 834] and lineage data [2819, 4076], as are haltere disc and 3rd-leg disc; and (6) genital disc spans A8–A10 [429, 735, 2343]. N.B.: H1–H3 are stylized [1739] and probably not contiguous [832, 1300, 1631, 2104, 4825]. See [2103, 3058, 3631] for head and tail details and [631, 3791, 3792] for other nuances. In the schematic drawing of the larva (above), discs are spread out and the midsection is omitted. See also App. 7.

mRNA [939] , and “Caudal” is made from uniformly distributed mRNA whose translation is repressed by Bicoid [3116] . Among these three factors, only Bicoid appears to be a bona ﬁde morphogen (see below) [1945, 2886, 4154, 4163]. The division of labor between maternal and zygotic (gap, pair-rule, and segment polarity) genes [2105, 3152, 4653] makes sense evolutionarily, given the time constraints of embryogenesis [4650]. 2. Gap genes. During the syncytial blastoderm stage, genes in the “gap” class are expressed in broad (∼10– 50% egg length) bands [1398]. They are so-named because LOF mutants are missing large parts of the segment series. Some initial overlaps between ex-

pression bands are erased by mutual repression [2329, 3567]. 3. Pair-rule genes. Spatial periodicity ﬁrst appears when genes of the “pair-rule” class are transcribed. Their name comes from an absence of alternating segments in their LOF mutants. This echelon, which came as a surprise [659, 3248], is divisible into primary and secondary tiers [1750, 3248]. Primary pair-rule genes integrate nonperiodic cues from axis and gap genes (with some inter se inputs), and most of their expression stripes (∼7 per gene) are separately regulated by a combination of upstream factors [242, 1722, 1750]. Secondary pair-rule genes rely more on the periodic outputs of primary genes. For example, fushi tarazu

78

IMAGINAL DISCS

Percent Egg Length

gap

90

80

70

60

50

40

30

20

10

bcd Bcd

0 Bcd

Cad

cad

Hb

kni

Kni

gt

pair rule

Kr eve ftz

Ftz

en wg Antp

homeotic

segment identity

100

hb

seg. pol.

segmentation gene hierarchy

axis

a

D

Ubx

A

P V

abd-A Abd-B

b Parasegments

1 2 3 4 5 6 7 8 9 10 11 12 13 14

Segments H4 H5 H6 T1 T2 T3 A1 A2 A3 A4 A5 A6 A7 A8 A9 FIGURE 4.2. Categories of genes involved in body segmentation and segment individuation.

a. Protein levels (Bcd, Cad, Hb) or mRNA levels (remainder) detected or inferred [3605] for a few members of each class in the segmentation gene hierarchy (axis, gap, pair rule, segment polarity), or (below) expression patterns of key homeotic genes at the time when their activity is instrumental in patterning [3248]. Gene abbreviations: abd-A (abdominal-A), Abd-B (Abdominal-B), Antp (Antennapedia), bcd (bicoid), cad (caudal), en (engrailed), eve (even skipped), ftz (fushi tarazu), gt (giant), hb (hunchback), kni (knirps), Kr (Kruppel), Ubx (Ultrabithorax), wg (wingless). Embryos at right depict realistic Bcd, Kni, ¨ and Ftz protein distributions [105, 661, 1102, 1103, 4055]. The Cad gradient arises via Bcd repression of cad mRNA translation [1487, 3116]. Vertical lines show a few other interactions: (1) boundaries of eve stripe 2 (arrowheads) are deﬁned by gt and Kr [4764]; (2) even-numbered en stripes (arrowheads) emerge due to a trailing gradient of Eve that turns off odd skipped (not shown) in the front cells of each ftz stripe (thus activating en by repressing its repressor) [877, 1324, 2681, 2682]. See [1605] for a comparable chain of negative regulators and [2002] for Deformed, where the control is more combinatorial than hierarchical. b. Embryos make 14 parasegments and 15 segments, with an en stripe (shaded) at the front or rear of each, respectively. Panel a is adapted from [162, 1416, 1946, 3605], with eve stripe 2 and en stripe 4 data from [4764] and [1324]; b follows [3843] (“A9” stands for A9–A10). N.B.: As a “wiring diagram,” this illustration is woefully inadequate because it omits so many key genes [47, 380, 1962, 3248], spatial details (e.g., uneven levels within homeotic gene domains [682, 2717]), and dynamic modulations [242, 1790, 2181, 3770, 4834]. Nevertheless, at least it conveys a feeling for the ﬂow of control. See [1487, 2121] for excellent exegeses and [2869, 4893] for evolutionary context.

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

has a compact “zebra” cis-enhancer that governs all 7 stripes [4834] . 4. Segment polarity genes. At the cellular blastoderm stage, the foregoing factors turn on the “segment polarity” genes. Their name comes from the segmental periodicity of LOF defects and associated reversals in cuticular polarity. Among them, engrailed and wingless are instrumental in setting metamere boundaries. Bicoid and Caudal are transcription factors [1011, 1709], as are all gap and pair-rule gene products [3768] , whereas Nanos acts on mRNA processing [4756]. These roles make sense because early mitoses are syncytial, which allows for free diffusion of signals among nuclei [1962]. By the time pair-rule stripes sharpen, however, plasma membranes are partitioning the cortical nuclei into a cellular blastoderm [47, 657, 3843], and direct internuclear communication is no longer possible [1429]. Not surprisingly, segment-polarity genes encode all sorts of proteins involved in transducing signals from the cell surface to the nucleus [3192, 3341]. Only two signals are critical here: Wingless and Hedgehog [1010, 1167, 4499] . Ultimately, the segmentation gene hierarchy regulates expression of “Hox” genes, which implement segmental identities (cf. Ch. 8 and App. 1) [2002] . Control of Hox genes stems mainly from the gap gene level, but inputs come from other levels as well (cf. Fig. 8.1) [683] . The overall circuitry is complex because many of the ∼50 core segmentation genes interact with ≥5 other such genes in the same or different echelons [1487, 2121] . Moreover, in some cases, the links skip a level or go back up the chain of command [3815] . Details are still being worked out [3383] , but the main design features are now clear [968] : 1. Both analog and digital controls are employed, and the driving inputs can be positive or negative [657, 2446] . 2. Binary (on or off) states are stabilized by positive feedback [2057] and arise, at the DNA level, via cooperative binding of trans-acting factors at cisenhancers of target genes [1708] . 3. Inputs are typically processed by combinatorial logic (as “and” gates) at these cis-enhancers [414] , while other aspects of Boolean syntax are mediated by competitive binding [3993] , dimerization [3768] , quenching [1600] , physical spacers [3992] , and “dual control” [657] .

79

These elements mesh to form a dynamic and robust system [4499] that orchestrates thousands of target genes [2541] . Despite its virtuosity, however, this “gene machine” is not optimized [4670] since certain pairs of components (nanos and hunchback [1945, 1986, 4154] or odd skipped and engrailed [878]) can be removed with little or no effect on the patterning of denticle belts. Aspects of the system have been modeled mathematically [2811, 3704], but computer models are best for tracking its behavior in real time. Simulations by John Reinitz [3567] and others [380, 2200, 2201] are yielding more testable predictions as additional parameters of the components become known [1010]. By 1987, the segmentation gene network was coming into focus, and seminal reviews were authored by Philip Ingham and Michael Akam [47, 1962]. Tellingly, these normally reserved British dons could barely contain their excitement at humankind’s ﬁrst glimpse into the molecular clockwork of pattern formation. Indeed, the insights galvanized the whole ﬁeld [125, 3669]. Akam realized that the embryo’s strategy of making segments from overlapping stripes fulﬁlled Curt Stern’s vision of prepatterns [47]: The pattern generated by the gap and pair-rule genes is transient and can appropriately be described as a prepattern.

As discussed in Ch. 3, Stern thought that prepatterns are transient networks whose interdependent nodes evoke elements in the ﬁnal pattern [4100, 4346]. In this case, the nodes would be the boundaries that delimit metameric regions along the A-P axis [657, 2433], and their interdependence would be attributable to the circuitry of the segmentation gene hierarchy [3248]. Ironically, Stern’s Prepattern Hypothesis had been toppled as the ﬁeld’s reigning paradigm in 1969 by Lewis Wolpert’s “Positional Information Hypothesis” [4724]. Wolpert argued that patterns are organized by global coordinate systems instead of by skeletal frameworks (see below). Akam’s remark heralded the resurrection of the prepattern concept as a useful way of thinking about embryos and discs [2446, 3953], although its full rehabilitation took several more years. To contrast these distinct schools of thought, the history of their conﬂict is brieﬂy recounted below. Although retrospective, the next section is not antiquarian, and readers would do well to study it. It constitutes the bedrock for the rest of the book.

80

Prepatterns and gradients clashed in trying to explain homeosis Stern’s reasoning was straightforward. If prepattern genes establish the prepattern’s nodes and the nodes are interdependent, then disabling one or more nodes within a LOF mutant clone should cause the remaining nodes outside the clone (in wild-type terrain) to redistribute themselves. In other words, any prepattern gene would have to behave nonautonomously in genetic mosaics [4100]. Today we know many such genes, most of which encode diffusible signals, but at that time (1954) none were known. Stern and his associates tested one mutation after another to see whether they could alter patterns nonautonomously in mosaics. By 1967, they had assayed ∼20 genes but had not yet found a bona ﬁde case of pattern reorganization [4346]. In 1967, Stern’s long quest for a “prepattern mutant” was rewarded when he and Chiyoko Tokunaga tested eyeless-Dominant (ey D ) in mosaics and found that it acted nonautonomously [4100, 4346]. Although its name comes from its eye defects (a result of cell death [1309]), this mutation also causes extra rows (∼2–6 rows vs. one in the wild-type) of sex comb bristles (∼39 vs. ∼10 bristles in the wild-type) on male forelegs. In mosaics, wild-type cells were found to be able to participate in extra rows when they intrude into the sex comb area of an otherwise mutant leg [4109]. Apparently, ey D augments the sex comb node or “singularity” in the leg prepattern, and wild-type cells respond accordingly – i.e., “ey D mutation bigger sex comb singularity phenotype” (“ ” denotes “causes”). However, other aspects of the mutant syndrome are visible in forelegs of both sexes and hence cannot be speciﬁc for the sex comb: swelling of the distal basitarsus, extra bristles of other types, disruption of pattern and polarity, and fusions of 1st and 2nd tarsal segments. These additional defects suggested other etiologies [1421, 1799]: overgrowth 1. Overgrowth Scenario: ey D mutation more sex comb positional coordinates phenotype [3213, 3448, 4348, 4724]. Conceivably, ey D induces overgrowth by disrupting Sex combs reduced [3332, 3333] – a Hox regulator at this spot. 2. Segmentation Scenario: ey D mutation gap in intersegmental membrane distorted gradient more sex comb positional coordinates phenotype [523, 2427, 3427, 4100, 4517]. This argument is predicated on the assumption that the ﬂy leg uses a sawtooth series of segmental gradients to specify positions along its

IMAGINAL DISCS

proximal-distal axis [385, 386, 1801, 2425]. Interrupting the barrier between the low end of one gradient and the high end of the next one (i.e., intersegmental membrane) could cause a “backﬂow” of morphogen, with the consequences listed above [2423]. 3. Apoptosis Scenario: ey D mutation cell death compensatory overgrowth phenotype [1314, 3441, 3442, 3964, 4346] (see [4336] for a striking phenocopy). Despite the fact that we still do not know which of the above explanations (if any) is right, ey D gave a ﬁtting ending to the prepattern saga, and Stern featured this story in his 1968 opus Genetic Mosaics and Other Essays [4100]. The celebration was short lived, however, because Wolpert proposed his model in the next year. The coup de grace that killed Stern’s model was not the paucity of prepattern mutants, which he rightly attributed to lethal side effects that would preclude their survival [4100]. Rather, it was the fact that homeotic mutations act autonomously [526, 1357, 4346]. “Homeotic” mutations, by deﬁnition, transform particular body parts to resemble other body parts – e.g., a leg into a wing (cf. Ch. 8) [2509, 3214, 4486]. Assuming that each organ has its own prepattern, Stern expected mosaic organs to display signs of jousting prepatterns – nonautonomous inﬂuences of homeotic tissue on nearby wild-type cells or vice versa [4097, 4098]. By 1968, however, several contradictory cases had been documented. For example, extra sex combs LOF partly converts 2nd and 3rd legs into 1st legs [1713], yet acts autonomously in mosaics [4349]. This result was not too unsettling for the hypothesis because all leg cells might “speak” a common language. No such rationalizing was possible, however, for bithorax LOF (anterior haltere into anterior wing) [2506] or aristapedia LOF (distal antenna into distal leg) [3444, 3614] because it seemed inconceivable that the same scaffolding could be used for organs as different as halteres and wings or legs and antennae (Fig. 4.3) [3214]. In his 1968 book, Stern acknowledged this paradox but did not offer a solution [4100]. In a review of Stern’s book, Peter Bryant argued that stretching the prepattern concept to cover these situations undermined its usefulness to the point of absurdity [520]. Some workers, including Stern, have employed the concept [of prepattern] as specifying, in Drosophila, individual bristle positions. However, experiments using mosaics for homeotic mutants tend to indicate that the prepattern for antennal structures is identical to that for leg structures. In that case, the prepattern could not be specifying individual bristle

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

positions since the bristle patterns are entirely different in these two appendages. If the concept must be generalized to such an extent that the prepattern for leg is identical to that for antenna, then it seems to lose much of its usefulness as a working hypothesis. We are left, in fact, with the rather hackneyed interpretation of pattern formation as resulting from a gradient of some hypothetical morphogenetic influence. [520]

A new conceptual framework emerged in 1969 when Lewis Wolpert blended Hans Driesch’s old notion of coordinate systems [1101, 3741, 3742] with modern information theory [95, 2547, 2766, 2989] to create the idea of positional information (PI) [4724, 4731, 4734]. It was not this homeosis problem that PI aimed to solve, but rather the “Regulation Riddle”: how can patterns robustly regenerate? Nevertheless, an explanation for homeosis followed naturally. Indeed, Wolpert cited the autonomy of aristapedia LOF as support for the “universality” corollary of his theory – namely, that the same PI coordinates should be readable by cells of any histotype. Coincidentally, in that same year Wolpert was handed an even better example. Antenna-to-leg homeosis in Antennapedia GOF (Antp GOF ) affects not just the arista but the entire antenna, and the patchiness of the transformation in nonmosaic mutant ﬂies revealed a striking correspondence between antennal and leg regions (Fig. 4.3a) [3445, 4728]. Such different organs could not possibly share a common prepattern (or so it was thought), but they might easily use a common coordinate system. The Prepattern Hypothesis seemed doomed. For me, the most significant contributions to the study of pattern formation over the last 30 years come from the work of Stern on genetic mosaics and the concept of prepattern. . . . This work provides excellent evidence for the concept of positional information and polarity potential, and the best evidence for the postulate of universality, at least within the same animal. As will be seen, the concept of positional information gets over some of the difficulties associated with the concept of prepattern. . . . Along similar lines one can interpret homeotic mutants which involve genes such as aristapedia which cause antennae to form legs. Once again the positional information may be the same and only the interpretation different. This is in line with the postulate of universality and it is thus again encouraging to find genetic mosaics of aristapedia with normal tissue behaving [autonomously] according to position and genome. [4724]

Wolpert invoked three stages for patterning and used a French Flag to abstractly represent three distinct cell types [4723]:

81

1. Speciﬁcation of PI. A scalar variable changes linearly or exponentially along an axis [3989] (e.g., the activin gradient [1656, 4220]). In the archetypal model, the variable is a “morphogen” (diffusible signaling molecule) whose concentration describes a “gradient” across the “ﬁeld” of cells whose fates it “speciﬁes” [4732]. Ambiguities in these terms can be minimized by operational deﬁnitions [967, 4673] (e.g., a “morphogen” is a signal that elicits expression of different genes at different concentrations [865, 1655]). Morphogens were imagined to emanate from a terminal source, with or without a sink where the morphogen is degraded [210]. Gradients are usually assumed to reach equilibrium before being “read” [3074], but this need not be true [2780, 3225]. Also unclear is whether cells “ratchet” to a particular level of response whence they are unable to retreat [1657]. In theory, cells can tell the absolute size of a ﬁeld and their orientation therein by sensing slopes of gradients [2434, 2448]. Finally, there is the issue of resolution (signal-to-noise ratio) as a function of distance. For example, in ﬂy eyes, the signal for ommatidial polarization is sensed over distances of >100 cells with 100% ﬁdelity (cf. Fig. 7.5) [2883], but how? 2. Recording of PI. To gauge its position, each cell measures the local height of the PI variable – like a person guessing the nearness of an ambulance by the loudness of its siren. (How ﬁnely cells can measure concentration differences is unknown [1142, 1655, 4366].) Cells then record this information (e.g., as on or off states of “memory” genes) [1608, 1964, 2417, 2515, 2803, 4730] as a function, perhaps, of receptor occupancy on the cell surface [1129, 1655, 2779] or of sensitivity thresholds in target gene promoters [1284, 1923, 2678, 3087, 3406]. The gradient itself can then disappear since it is no longer needed. How long such memories can persist in dividing cells is unclear. 3. Interpretation of PI. At a later time, cells translate the memories of their old locations (“positional values” [4729, 4730, 4732, 4734]) into differentiated states (blue, white, or red) through some sort of downstream gene circuitry [98, 2646] – like a letter carrier deducing city names from postal codes (e.g., see [472]). Whereas Step 2 converted analog to digital information, Step 3 must convert quantitative to qualitative states [1655], although memories may also control analog traits such as adhesivity [2436] or mitotic rate [1263]. One abiding mystery is the extent to which neighboring cells interact to sharpen the interpretation zones [1143, 1609, 4691].

IMAGINAL DISCS

antenna

xi m

ro

p

A2 A3 Co

T1

b Models

Fe

leg

Positional Information Hypothesis 7 6

Gra

dien

4 3 2 1

1. Specification

t

? ? ? ? ? ? ? subregions 2. Recording

7 6 5 4 3 2 1 3. Interpretation

(A-type)

positional values

(L-type)

pattern

Ar BC

A2 A1

A3

antenna

x T5 T2 T1 Ti Fe Tr Co Cl T3 T4

leg

Prepattern Hypothesis 1. Prepattern 2. Competence

singularities

x (A-type)

(L-type)

pattern

x

antenna

leg

xi m

D

Ti

[M] 5

V distal

T4 T3 T2

Tr

ro

BC T5

V

al

Cl

A1

Ar

distal

a Homology map

D

al

82

p

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

PI and prepatterns are often contrasted in terms of how they might create an array of evenly spaced bristles within a line of cells [526, 2764, 4671]. PI would use a gradient to assign a number to each cell, and certain numbers would signify “bristle” because the genome is “wired” accordingly. A prepattern strategy, on the other hand, would use some sort of device (e.g., reactiondiffusion) to put bristle-inducing signals at regular intervals, and any cell would be able to respond (above a certain threshold). With PI, every cell knows where it is and differs from every other cell, whereas prepatterned cells would know their state (bristle vs. nonbristle) but not their location [4725, 4726, 4729].

83

In the PI scheme, patterns are essentially encoded as bitmaps [1805, 3610], with each cell (≈pixel) having a separate genetic representation (≈on/off state). The huge coding capacity needed for intricate patterns (even with combinatorial encoding) strained credulity [862, 2142, 2646], as did the paucity of plausible molecular mechanisms [1805, 2809, 3087, 4730, 4732], but neither quibble slowed the PI model’s rise to popularity [864, 3988, 4734]. It quickly attained paradigm status. Stern’s camp formally “surrendered” in 1978 [4346]. Soon thereafter, direct support for PI came from the segmentation genes [4733]. They furnished the ﬁrst deﬁnitive morphogen (Bicoid) [1103] , plus insights into

FIGURE 4.3. Demise of the Prepattern Hypothesis in 1969 and ascendancy of the Positional Information Hypothesis, due in part to their clumsy or elegant explanations for homeotic autonomy. a. Correspondence (arrows) between antennal and leg parts, as deduced from Antennapedia R (Antp R ) ﬂies, whose antennae are patchily transformed into 2nd legs [3445]. Abbreviations: A1–3 (antennal segments 1–3), BC (basal cylinder), Ar (arista), Co (coxa), Tr (trochanter), Fe (femur), Ti (tibia), T1–5 (tarsal segments 1–5), Cl (claws). Compasses denote adult polarities (D = dorsal, V = ventral; the inversion is explained below). Antennae of Antp R ﬂies (not shown) have patches of leg tissue. Whenever leg tissue arises at a particular proximal-distal level, it makes leg structures appropriate for that level. The implication is that these leg cells can somehow ascertain their location along this axis of the antenna. Wolpert and Stern agreed that the antenna and leg probably share some kind of “ground plan” that both types of cells can “read.” They disagreed over the kind of ground plan. In Wolpert’s scheme, it is a gradient; in Stern’s, it is a prepattern. b. The two models. Each of the 7 gray rectangles represents a segmental region (eventually containing 102 –103 cells) of the antennal or leg rudiment along the proximal-distal axis. Certain segments are subdivided or grouped. Lewis Wolpert’s Positional Information (PI) Hypothesis explained antenna-to-leg homeosis in terms of a shared coordinate system [4724]. Antennal and leg cells are supposed to use the same positional signal – a diffusible molecule “M.” This “morphogen” is produced at one end of the organ (left) and diffuses to form a concentration gradient (triangle). The gradient could be linear (as shown) but is more likely exponential. In Step 1, cells assess their distance from the source by measuring the amount of M: the more M that a cell “tastes” at its site, the nearer it “thinks” that it must be to the source. In Step 2, cells record these levels (1–7) as “positional values” (on or off states of memory genes?) that persist after M vanishes. In Step 3, cells translate these values into structures that suit the disc to which they belong. For example, antennal cells interpret the number 7 as “arista,” whereas leg cells interpret it as “T5 and claws.” The interpretation mode that a cell uses (A vs. L) is predetermined by the turning on or off of a particular homeotic gene in one of the cell’s ancestors. Mutations in that gene cause homeosis: mutant cells “think” they are leg cells, but they can still sense M properly, so they make leg structures that are appropriate for their proximal-distal level in the antenna. Finer-grain patterning within subregions may rely on a second echelon of (segmental) gradients (not shown) [386]. The elegance of Wolpert’s model has been its universality: the same gradients can theoretically pattern all the discs [20, 523, 3701]. Curt Stern’s Prepattern Hypothesis had a more difﬁcult time explaining homeosis. Its basic assumption was that every pattern element is preceded by a discrete signal (“singularity”) [4095]. Unlike the morphogen, these signals (circles, ovals, square, triangles, and “x”) differ qualitatively: they are not graded. In Step 1, the disc lays out an array of these signals (the “prepattern”). The autonomy of the Antp R mosaicism implied that antenna and leg must express one another’s signals. (How else could nouveau-leg cells that ﬁnd themselves “stranded” in the antenna ﬁgure out what structures to make?) In other words, both prepatterns must coexist in each disc [4346]. Antennal cells would normally read one subset of signals (white symbols), while leg cells read the other (black symbols). The ﬁltering of these subsets (cf. Fig. 3.2) would depend on a cell’s “competence,” which in turn would be determined early in development when one of the cell’s ancestors turned a particular homeotic gene on or off. Mutations in such genes would lead to interdisc transformations (cf. Ch. 8). This argument was tolerable for a single type of transformation (e.g., antenna-to-leg), but the large number of interdisc homeoses [3214] implied that every disc must contain the prepattern for virtually every other disc. Although not impossible, this notion seemed implausibly clumsy, especially when contrasted with how easily the PI Hypothesis handled this same phenomenon. Panel a is redrawn from [3447], which depicts a right antenna and left leg (see [1587] for A3 subregions, [1516, 1561, 2287] for molecular “homology,” and [4522] for a different map). Panels b and c are based on the ideas of Stern (Fig. 58 in [4100]) and Wolpert [4734], as contrasted inter se by Bryant [526] and Tokunaga [4346]. Black or white symbols in the antennal and leg schematics are merely abstract representations of the actual pattern elements. See Figure 8.3 for data on how antennal vs. leg identities are actually controlled and [2956, 3744] for the origin of the gradient concept. See also App. 7.

84

(1) how gradients can specify zones (e.g., gap gene enhancers varying in Bicoid-binding afﬁnity) [1105] , (2) how boundaries can be sharpened (e.g., competitive and cooperative trans interactions) [575] , and (3) how interpretation can be implemented (e.g., morphogen sensitivity being dictated by distance from cis-enhancer to core promoter) [1835, 2421]. The first step involves the localization of a cytoplasmic determinant, in this case the messenger RNA of . . . bicoid in the unfertilized egg. On fertilization the localized RNA becomes translated into bicoid protein, which spreads by diffusion and forms a morphogenetic gradient. The bicoid protein concentration provides positional information to the nuclei in which it is taken up. The third step involves the conversion of the graded pattern into a repetitive pattern of stripes under the control of the segmentation genes. Finally, the repetitive pattern is converted into a sequential pattern; the segments differentiate and each acquires its own identity [via] the homeotic genes. [1416]

Gradients and threshold responses appear to be used not only in Bicoid’s control of gap genes [1284] , but also in cross-talk among gap genes [4764] , in gap gene control of pair-rule genes [1323] , in cross-talk among pair-rule genes [4553] , and in pair-rule gene control of segment-polarity genes [1324] . Despite the network’s reliance on gradients, the way that it processes information is more reminiscent of prepatterns [2448]. To wit, cells only use certain Bicoid levels to establish zones of gap gene expression, and the zone borders act like singularities to delimit pair-rule stripes. The remaining levels are ignored. In a pure PI scheme, all levels should be recorded as heritable positional values [3087], but Stage 2 of the PI scenario seems to be bypassed. Thus, the process seems more like a PI-prepattern hybrid than strictly one or the other. Peter Lawrence, like Michael Akam, realized that these theories could dovetail nicely into a uniﬁed explanation [2446]: [The prime function of pair rule genes is] to locate boundaries that delimit fields or gradients of positional information. [2433]

Within each segment, cell positions were thought to be speciﬁed by a “segmental gradient” [2426] , based on a long history of surgical experiments [1300, 2430] . The segment-polarity echelon fulﬁlled this prophecy insofar as its stripes have a segmental periodicity [2717] , although the initial metamerism consists of “parasegments” [1978, 2438, 2724] that are out of phase with segment boundaries (Fig. 4.2b) [1772, 4152]. Two protein products of segment polarity genes – Hedgehog and Wingless – act like PI morphogens in

IMAGINAL DISCS

embryonic body segments [1789, 2448, 2716, 3748] and adult abdominal segments [2301, 2303, 3905, 4157, 4158] (see [2649] for a prescient model). Early experiments seemed to refute Wingless as a morphogen in the embryo [3730], but this conclusion was overturned by later studies which showed that sloppy paired imposes an uneven “landscape” of cellular competence to respond to Wg [596, 3129].

Homeotic genes implement regional identities How could identical gradients (segmental or parasegmental) produce different structures? According to PI theory, they must use different “code books” for interpreting their coordinates. In terms of Wolpert’s ﬂag metaphor, each segment would use the same coordinates, but Segment T1 might employ a British Flag interpretation mode, while Segment T2 uses an Italian Flag mode, etc. Only a few bits of information would be required to establish 15 or so interpretation modes along the A-P axis (one per segment). The process whereby segments become different from one another (“individuation”) is genetically separable from segmentation, because certain homeotic mutants undergo normal segmentation with virtually no individuation [2507, 4147]. What would happen if the interpretation mode of a particular segment were to malfunction? The segment should develop like another segment (i.e., undergo homeosis), and the same is true for interdisc transformations [1418]. Thus, the phenomenon of homeosis ﬁt neatly into the PI paradigm [523, 3701]. Presumably, homeotic genes would serve three PI-related functions: (1) use their on or off states to record various levels of the whole-egg gradient, (2) perpetuate these memories during disc growth, and (3) set the interpretation mode for each segment or parasegment [1365, 2755, 3882]. These expectations turned out to be basically correct (cf. Ch. 8). As Wolpert’s ideas were rippling through the research community in the early 1970s, a complementary model of Stuart Kauffman’s was also gaining notoriety [2155, 2156]. Its formulation for disc identities was based on fate switches observed during “transdetermination” – a sporadic metaplasia seen during long-term culture of disc tissue in vivo [1668--1670] or in vitro [3883, 3885]. Kauffman was struck by the fact that each disc only transforms into a few other types (cf. Fig. 6.9d). He argued that (1) these limitations are indicative of an identity code, (2) the code is binary, and (3) the “transition rules” (i.e., which disc can transform into which other discs) apply equally well to homeosis [2153, 2164, 3214]. According to

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

his “Binary Code Conjecture,” each disc follows a preferred sequence of transdeterminations because only one bit changes at a time (i.e., 0 to 1, or 1 to 0). For example, if the code for 2nd leg is “1110,” then single switches could produce 0110, 1010, 1100, or 1111, but not other states (e.g., 1001) [2158, 2159]. Transdetermination and homeosis share other features [2082, 2159, 2754, 3788, 4142], although differences do exist [2138, 3448, 3883], and there are other possible explanations aside from coding (cf. Ch. 8) [1472, 2411]. In 1975, Antonio Garc´ıa-Bellido proposed his “Selector Gene Hypothesis” [1358]. Like Kauffman’s conjecture, it invoked switch genes for regional identities [701, 1635, 2930, 3457], but the regions were lineage compartments within discs rather than entire discs [2431]. At the time, it seemed conceivable that a succession of compartments might subdivide discs down to the single-cell level [1358, 2142, 2440], so that each cell would acquire an “area code” tantamount to a Wolpertian coordinate [1369, 1477, 3914]. However, that idea proved false [354, 494, 1375, 1639, 4671]. The discovery of the homeobox, announced by several labs in 1984 [2783, 2785, 3844], provided dramatic support for the selector gene idea [1412--1414, 2421]. This ∼180 b.p. sequence encodes a ∼60 a.a. DNA-binding domain (cf. App. 1) [1417, 1420]. As homeoboxes kept showing up in one cloned homeotic gene after another, it became clear that these genes constitute a distinct class structurally and functionally [1418, 3556] as contrasted with “housekeeping” genes whose products (e.g., metabolic enzymes) are used ubiquitously [293, 1865]. Since then, more evidence has surfaced for “master genes,” but their roles defy all the early models. Because it is not possible to understand how homeotic genes act without delving into each disc’s idiosyncrasies, further discussion is deferred until Ch. 8. One quirk of homeosis itself, though, merits mention here – the “Collective Amnesia Conundrum” [1805] (cf. “homeogenetic induction” [1005] and the “community effect” [4063]). Most models presume that disc states are carried by individual cells. A leg disc cell, for example, would know that it belongs to a leg disc and not to some other disc, regardless of its environment. This axiom was based on autonomy in genetic mosaics (see above) [2922, 4346] and retention of histotypes by dissociated disc cells in mixed aggregates [1374, 1411, 1421, 3138]. However, the islands of leg tissue in Antp GOF antennae arise nonclonally [3445], as if neighboring cells jointly forget their disc of origin, and transdetermination involves the same sort of process [1353, 1405, 1406]. Social cell behavior like this might depend on the nonclonal clus-

85

ters of synchronously dividing cells that are often seen in normal discs [23, 532, 2744, 2848] if (1) disc cells normally forget their identity at some stage of the cell cycle [4882], (2) they recover a sense of who they are by “reading” the identities of neighbors when they exit mitosis [533], and (3) too many cells simultaneously undergo mitosis, in which case groups of cells would remain amnesic and be forced to adopt a default state. Excess growth is indeed correlated with transdetermination [1421, 1670, 3883, 4334] and with “homeotic regeneration” [1421, 3448, 4142], but this “Cyclic Amnesia Scenario” is too simple [3448] because it predicts 1. Only one default state should exist per disc; however, each disc can actually transform into several other types [1421, 3214, 3881]. 2. Tumorous discs should exhibit frequent fate switches in situ; however, they typically do not [541, 1386, 2107, 3887]. 3. Fate switches should occur in normal discs by chance alone. On the contrary, homeotic outgrowths are virtually never seen in wild-type ﬂies [3883]. Observed cases of communal conversion may arise from diffusion [4063, 4563]: if single cells mistakenly start expressing a morphogen, then it could diffuse and activate inappropriate gene circuits so that unrelated neighbors jointly switch their fate [2082, 2754] (cf. Ch. 8). The dependence of transdetermination on wounding is consistent with this notion [3887] because wounding can evoke morphogens [489], and some homeoses are associated with tissue loss [839].

Wing and haltere discs ‘‘grow out’’ from 2ndand 3rd-leg discs Rudiments of the thoracic discs become morphologically recognizable when they invaginate after germband shortening (St. 13) at 9–10 h AEL (After Egg Laying, 25◦ C) [237, 2717, 2820]. At this time, the wing and haltere discs contain ∼24 and ∼12 cells, respectively [237, 664, 827, 2335], excluding mesoderm cells [57, 1886, 2819]. In a newly hatched larva (∼24 h AEL), the major discs contain 20–70 cells (wing ∼40; eye ∼70; leg ∼40; haltere ∼20; genital ∼60) [2650]. In mid-to-late 1st instar, all disc cells resume mitosis [474, 528, 2650] (they had stalled at ∼10 h AEL, St. 13 [1142, 1259]) and start multiplying exponentially [3139, 3441]. During the 3rd instar, cell numbers double every ∼10 h (wing 8 h; eye 11 h; leg 8 h; haltere 13 h) [2081, 2935], so that by pupariation each disc has 10–50 × 103 cells (wing ∼49,000; eye ∼44,000; leg ∼17,000; haltere ∼10,000) [542, 1123, 1185, 2710] and measures 50–300 µm across [524, 3139, 3507]. Indeed, one reason for

86

studying discs is that they must pattern themselves during growth – a constraint that is typical of most developing systems (e.g., vertebrates) – whereas ﬂy embryos employ a quirky syncytial phase whose circuitry may be less useful in trying to understand common (ancient) patterning mechanisms [1303, 3074, 3747]. When do disc cells become determined? That is, when do they irrevocably decide to make (1) adult vs. larval structures [1717] and, more narrowly, (2) structures appropriate to their own disc type vs. some other type [523, 1411, 2411, 3139, 3795]? Both decisions must occur after 3 h AEL (St. 5) because single blastoderm cells can contribute to both imaginal and larval tissues when transplanted homotopically at that time [2819]. Leg determination probably happens by 5.5 h AEL (St. 11) because the early enhancer for Distal-less (Dll) is activated then [827, 1571, 1572]. Dll is a marker for distal leg, although its expression is not so restricted initially [1572]. The ﬁrst transcription of vestigial (vg), a marker for wing and haltere cells, is not detected until ∼2 h later [827, 4681]. Strangely, however, neither Dll nor vg is required for disc development at this stage [827, 3969, 4681], so neither gene provides ironclad evidence for determination [884, 4682]. Nevertheless, the ∼5–7 h AEL (St. 11) estimate is probably close [2335], given that (1) disc fates can be changed until that time (but not afterward) by misexpressing homeotic genes [684, 3930] and (2) X-rays can induce pattern duplications in discs (implying that they have become embryonic ﬁelds) after 6 h AEL (but not at 3 h AEL) [4651]. Remarkably, the Vg-expressing cells in Segments T2 and T3 originate as subsets of Dll-expressing cells that move dorsally in the extended germ band [827, 1571, 1572]. Because this movement is correlated with locally increased mitosis [237], it is probably driven by growth, although cell migration is likely involved later when wing and haltere cells cling to tracheal branches [237, 2335]. Evidently, wing and haltere progenitor cells arise within the same nests that form 2nd and 3rd legs, respectively (Fig. 4.4). This conclusion ﬁts with the visible “budding” of prospective wing from 2nd-leg rudiments (and haltere from 3rd leg) in the dipteran Dacus tryoni [81], and it afﬁrms the controversial idea that the insect wing evolved as a branch from the arthropod leg [145, 2348, 2349, 3911, 4662]. Also, it clariﬁes why single blastoderm cells can contribute to both wing and 2nd leg (or to haltere and 3rd leg) [2442, 4076, 4651]. Moreover, this recent ﬁnding settles an old dispute between two titans from the Morgan era: A. H. Sturtevant and Curt Stern. In his 1929 gynandromorph study (cf. Ch. 1), Sturtevant noticed that sternopleu-

IMAGINAL DISCS

ral bristles are clonally linked with thoracic bristles as often as with leg bristles [4180]. From this result, he inferred that the sternopleura (a ﬂank sclerite; cf. Figs. 4.4i and 5.1d) can come from either the wing or 2nd-leg disc, depending on which disc happens to spread into this region ﬁrst during metamorphosis (cf. other claims for indeterminacy [3250] that were refuted by Stern [343, 3006]). Stern argued (correctly) that sternopleurae always come from 2nd-leg discs, and he marshaled much evidence to prove his point [4090, 4099]. Sturtevant remained stubbornly unconvinced [4182]. To explain Sturtevant’s data, Stern guessed that leg and wing primordia must abut one another in the ectoderm [4090, 4652], and Dll and Vg now validate his 1940 conjecture: If the division line between [male and female areas] of a developing gynandromorph has an equal chance of falling either between the ventral [leg] and dorsal [wing] anlage or between the two subregions [sternopleura vs. remainder] of the ventral [leg] anlage, then the results of Sturtevant can be explained without recourse to the hypothesis of indeterminate overgrowth of imaginal disc ectoderm during metamorphosis. [4090]

By ∼7 h AEL (St. 12) when disc fates appear to be set, the leg and wing primordia contain ∼20 and ∼30 cells, respectively [827, 828, 2819] (including mesoderm cells [1886]), and these ﬁgures agree with indirect estimates from cell lineage studies [2828, 3139, 4649], which put the number of prospective epidermal cells per disc in the 10–40 cell range (excluding labial and clypeolabral discs). Because discs are not clones (cf. Ch. 1), the discinitiating factors (whatever they may be) need not be so precise as to pinpoint single cells in the embryonic ectoderm. Discs retain their identities during larval life [2411], although their cells look embryonic and do not differentiate until metamorphosis [82, 526, 3795, 3797]. Thus, it is not possible to distinguish wing- from leg-disc cells, for instance, just by looking at them [3165, 4424]. This prolonged period of “determination sans differentiation” [1654, 3062, 3448] made it possible to study the stability of determined states [1384, 1406, 1668] and growth regulation [543], which led to the discovery of transdetermination and regeneration, respectively [2755, 3794]. The independence of larval from imaginal development is vividly illustrated by discless larvae, which feed and grow and molt apparently normally and only die when attempting to undergo metamorphosis [3428, 3880, 3886, 4221]. Disc cells remain diploid [499, 1201, 1764, 1765, 2151], whereas the larval skin cells that surround them become polytene [1126, 1259, 1346] via endocycles of DNA replication

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

[1261, 3551, 3927, 3995, 4587].

Historically, differences like these made sense in terms of the once-popular view that separate batteries of genes control imaginal vs. larval development [1384, 4686].

To my mind the simplest explanation of such a clear cut dichotomy brought about by a single kind of stimulus is to be found in the analogy of the locked door to which the juvenile hormone is the key. In other words we are concerned with two sets of genes so disposed that in the presence of a small amount of juvenile hormone one set takes precedence, while in the presence of a large amount of juvenile hormone the other set is brought into action. [4659]

That view was primarily buttressed by “disc-speciﬁc” mutations recovered in screens for pupal lethals [3880, 3881, 3886, 3888, 4121] (especially ones that cause a discless phenotype), but subsequent work showed that defects can be disc-speciﬁc for a trivial reason -- viz., the egg has enough maternally supplied gene products to build the larva (in 1 day) but not the adult (requiring 4 more days) [834, 4221, 4257, 4671]. Regardless of whether the “Battery Dichotomy Hypothesis” has any validity in other areas (e.g., physiology), it is certainly defunct in the realm of patterning: many of the same genes that pattern the larval epidermis also pattern discs [834, 3341, 4675] and histoblast nests [4158]. Except for the humeral disc (which lacks a lumen), discs become hollow sacs via invagination or delamination [1421, 1544]. The apices of their cells face the lumen [142, 2650, 4424] and rufﬂe to form a carpet of microvilli prior to cuticle secretion [194, 3421, 3422]. The disc epithelium is one-cell thick [3165, 3426, 3539] like the ectoderm whence it comes [1259] and like insect skin in general [2584, 3421, 4661]. One side of each disc thickens into a columnar epithelium that will secrete adult cuticle [1133, 1315, 1431, 3564], while most of the other side forms a thin “peripodial membrane” whose squamous cells make few cuticular parts in the thorax [500, 2862, 2863] but signiﬁcant parts of the head [1777]. Curiously, mitoses may be synchronized across the lumen [2744]. Coordination of this sort (as well as other sorts of signals) is probably mediated by the ﬁlopodia (“translumenal extensions”) that stretch from peripodial cells to the columnar surface [773, 1473, 3508]. During 3rd instar, the columnar epithelium looks “pseudostratiﬁed” because its nuclei occupy many levels [2650, 4715]. In early 3rd instar, the wing portion of the wing disc (as distinct from the notal part) is initiated via a separate genetic circuit [885, 2219, 2254, 4683, 4684]. Characteristic folds emerge there (eventually forming a “pouch”) and in the leg disc (eventually forming a comparable “endknob”) [142, 377, 2287], presumably due to region-speciﬁc mitotic

87

rates or orientations [544, 2848, 3422, 4427] or histotypic afﬁnities [351, 1060, 2570]. Cell death cannot play a major role in morphogenesis or patterning [532, 3422] because it is so rare [12, 1501, 2015, 2848, 2849], except in eyes [396, 397, 489, 1309, 1763] where it serves to tighten the ommatidial lattice [4713, 4715]. Aside from epithelial cells, most discs contain adepithelial cells [239, 476, 497, 3104, 3660] (= mesodermal myoblasts [236, 1214, 1886, 2820, 3609, 3661]), neurons [4330], and tracheal cells – all of which reside between the epithelium and the basal lamina [3422, 3426]. Genetic evidence suggests that adepithelial cells in leg discs induce patterning in the tarsal epithelium [3332, 3333]. The male genital disc is freakish insofar as it recruits mesodermal cells into the epithelium [38]. Those cells go on to form the paragonia and vas deferens. During metamorphosis, each disc everts through its stalk [1311, 3422, 4429]. Historically, the planarity and insularity of discs (the latter feature being uncommon among embryonic ﬁelds [4516]) proved useful for analyzing mechanisms [523, 2429, 3448], as did the durability and intricacy of the cuticle, which indelibly records the ﬁnal state of virtually every epidermal cell [526, 3421, 4663].

Thoracic discs arise at Wingless/Engrailed boundaries Although blastoderm cells are not restricted to larval vs. imaginal fates [2819], nor to ventral (leg) vs. dorsal (wing/haltere), nor even to left vs. right disc fates [4076, 4651], they are conﬁned to anterior vs. posterior fates within each segment [4076]. Why do A/P compartments exist? One early clue was that A/P boundaries precede disc formation, so they might help decide where discs arise [2812, 3135]. In fact, they do. The wing’s A/P border traverses a featureless ﬁeld of hairs (Fig. 4.4) [354, 1376] – a ﬁnding that seemed odd because PI axes were thought to be “seams” for anatomic designs [904, 1409]. This invisible line was previously revealed in the phenotypes of bithorax LOF , postbithorax LOF , and engrailed1 mutants [4671]. The ﬁrst two mutations reside in the cis-regulatory region of Ultrabithorax (Ubx) [254, 3940] and hence are written more correctly as “UbxbxLOF ” and “Ubx pbxLOF ,” whereas en1 is an idiosyncratic allele of engrailed [2356] that does not behave as a simple LOF [852, 1138, 1636, 1837, 2447]. UbxbxLOF and Ubx pbxLOF transform the A or P part of the haltere into the corresponding part of the wing [2924] and A or P 3rd leg into 2nd leg [4324], whereas en1 transforms P into A compartments in the wing [493, 1378, 2306, 2441] and foreleg [439, 852, 4343]. Thus, while the Ubx alleles might be acting via interdisc

88

IMAGINAL DISCS

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

serial homologies, en1 cannot be doing so. Nevertheless, all three alleles ﬁt the Selector Gene Hypothesis since they imply an abstract (histotype-independent) code of some sort [1355, 1375]. Because en is on in the P part of each disc and off in the A part [495, 1697, 1963, 2307], it could serve as a PI “mode selector” gene to allow A and P cells to use different interpretation modes for reading the same PI gradient [2441,

89

2930]. If so, then the A and P gradients should be “back to back” [904, 4734], given the mirror-image phenotype of en1 in the wing and leg. Comparable duplications caused by other mutations are attributable to cell death followed by compensatory growth [523, 526, 539, 1499, 2015, 3441], but en1 cannot be acting thusly because it mainly behaves cell autonomously [1378, 2441, 2928, 4343]. Indeed, its autonomy led to Morata and Lawrence to postulate that

FIGURE 4.4. Initiation of thoracic discs at the Wingless/Engrailed interface and the later role of the A/P boundary during disc development. Gene abbreviations: dpp (decapentaplegic), Dll (Distal-less), en (engrailed), hh (hedgehog), vg (vestigial ), wg (wingless). a, b. Magniﬁed view of the embryo’s T2 (second thoracic segment, ﬂank area only) at two stages after egg laying (AEL). Axes (A-P, anterior-posterior; D-V, dorsal-ventral) are indicated by compass at left. a. T2 at ∼5 h AEL (extended germ band, St. 10/11). The posterior two rows of cells (hexagons) transcribe wg or en; key at right) and secrete Wg or Hh. Wg and Hh diffuse (black triangles = concentration gradients) and sustain one another’s synthesis by a feedback circuit [2717] that is abridged here (inset at right; see Fig. 2.7 for key and [2816] for theory). Because the gradients are symmetric, each should create two ﬂanking counterparts, but the response is asymmetric due to other genes (not shown) [596, 1628, 3129, 3748]. Before the stage in b, all ectoderm cells divide at least once. The pedigree (lines connect mother to daughter cells) depicts mitoses in one ﬁle of cells along the A-P axis, although divisions are not really so orderly. The important point is this: whether a cell expresses Wg or En is not heritable. To wit, as cells move away from the Wg/En interface, they stop making En [4491] and Wg [3363] (fading letters). b. T2 at ∼8 h AEL (germ band shortening, St. 12). Wg and En domains do not grow in proportion to segment width (although En stripes widen more than depicted [272, 1064, 2717]) because diffusion ranges of Wg and Hh are ﬁxed [1063]. An A-P stripe of dpp-on cells crosses a gap in the Wg stripe. Dpp is also a morphogen (gradient at right). Around the tip of the Wg stripe remnant, cells transcribe Dll [834]. The upper limit of the Dll-on spot is set (at ∼St. 11) by Dpp (Dpp Dll). The lower limit is set by Spitz (Spitz Dll) – another morphogen (not shown; cf. Fig. 6.12) from the V midline [1572, 2335, 3536]. c–e. Subsequent changes in the T2 Dll cluster. Reshaping is apparently due to mitosis (vs. active cell migration) [827, 1571], but cells also rearrange to some extent [1572, 2335]. c. Note the new symbol key. d. As cells move dorsally with the Dpp stripe (see e), they stop making Dll [1571] (fading letters). e. Eventually a dorsal cohort starts transcribing vg – a wing (and haltere) marker. Thus, wing and leg anlagen arise from a common pool of cells [827]. The pool is partitioned (St. 15) based on how much Dpp (D signal) the cells perceive relative to Spitz (V signal; not shown) [2335]:

High Dpp /Spi ratio Low Dpp /Spi ratio

“wing” state (vg-on/Dll-off) “leg” state (vg-off/Dll-on)

f–i. Mature (3rd-instar) discs and their adult derivatives, showing regions that do (black) or do not (gray) make En (oriented as per compass in a). Within both discs, the en on/off boundary becomes a reference line for specifying cell positions along the A-P axis, using either Wg or Dpp as a morphogen (not shown). In the leg disc, a remnant of wg-on cells is retained ventrally, and a stripe of dpp-on cells is induced dorsally (cf. Fig. 5.4). In the wing disc, a stripe of dpp-on cells is induced along the entire boundary (cf. Fig. 6.3). Plusses mark the appendage tips, which would come out of the page during eversion and then ﬂop down. f. Left wing disc. Tissue in the translucent rectangle (D wing surface) is not visible in g because it folds behind (cf. Fig. 6.1). Eversion brings the notum (NP, PA, and SC = notopleural, postalar, and scutellar areas) into contact with the pleura (pl). g. Adult wing (V surface only) and heminotum in side view. Bristles are omitted except at wing margin. A and P denote lineage compartments. The A edge of the en-on region per se is actually closer to vein III due to a shift during late 3rd instar (cf. Fig. 6.7d) [350]. h. Left 2nd-leg disc. i. Adult leg, drawn as if ﬁlleted along the ventral A/P boundary and spread ﬂat so its whole surface is visible. Thoracic structures also come from the leg disc, including StPl (sternopleural sclerite). Holes in joints are cartographic artifacts due to odd shapes of proximal leg segments. Panel a is based on [1063]; b–e are adapted from [834, 884, 1571]; f is traced from [2467]; g is modiﬁed from [1376, 4076]; h is sketched from stained discs pictured in [495, 2754]; and i is amalgamated from [1800, 2449, 4076]. N.B.: Cells are not usually packed so neatly [1064], nor are they so constant in shape or size [4743]. Indeed, they may not even stay in a monolayer [1572]. In b, dpp transcription is actually greater posteriorly along its stripe (possibly due to upstream control by segment-polarity genes) [2003]. In c–e, a ring of escargot-on cells (not shown; future pleura and coxa) surrounds the Dll-on core (all other leg segments) [1542, 1572, 1573]. Unlike the leg disc whose axes persist from the blastoderm, the wing disc appears to rotate ∼90◦ during development (not shown) [4652]. As shown in h, the A/P (en-on/off) line of the leg disc is diagonal near the center but zigzags toward the stalk above [495, 3747, 4254]. This irregularity has led to confusion because some authors schematize the A/P line vertically (D pole at 12 o’clock) like the outer part, while others slant it (D pole at 1 or 2 o’clock) like the central part (cf. Fig. 5.1).

90

“the state of activity of the selector gene is permanent and . . . propagated by cell heredity” [2440], and the discovery that en has a homeobox [1031, 1247, 3429] corroborated this conjecture [354, 1979]. Subsequent research, however, showed that at certain stages a cell only expresses en as long as its neighbors force it to do so [1839, 1968, 3310, 4491]. In general, the following terms help distinguish between these autonomous (independent) and nonautonomous (dependent) states [523, 3448, 3795, 3796]: 1. “Determined” states are intrinsic, permanent, and heritable [1411, 1421, 2448]. They are “ﬁrm biases” [3794]. 2. “Speciﬁed” states are extrinsic (imposed by intercellular signals), transient, and not heritable [2111, 2716, 4645]. They are “transitory biases” [3794] that depend on the persistence of the external signals. En and Wingless (Wg) stripes ﬁrst appear at the cellular blastoderm stage (∼3 h AEL, St. 5) [67, 176, 271, 2146, 3157]. When a dye was injected into single cells and their descendants were examined after several mitoses (∼5– 8 h AEL), 15 (out of 37) clones were found to contain a mixture of En-expressing and nonexpressing cells [4491]. Clearly, En states are not irrevocably ﬁxed in the blastoderm because if they were, then only pure (en-on or en-off) clones should exist. A second major ﬁnding in this study was that none of the 15 clones straddled a Wg/En (parasegment) boundary [1963]. The inability of blastoderm clones to cross this line conﬁrmed cell-lineage studies of leg and wing discs [2432, 4076] and showed that A/P lineage restrictions apply not only to discs, but also to the ectodermal fabric from which they are cut [2444, 4488]. In theory, segregation of A and P cells could be enforced by killing trespassers, but it seems that cells never even venture into alien territory. The reason appears to be an incompatibility of cell afﬁnities [2437]. That is, the en-on state endows a cell with (as yet unidentiﬁed) surface molecules (cadherins? [3308]) that prevent its immersion in a ﬁeld of wg-on cells, and vice versa. The idea that disc cells seek “their own kind” came from old experiments where cells from marked discs were mixed together, and the aggregates were found to form chimeric patterns. The harmonious integration in these patterns was attributed to homophilic cell sorting [1356, 4423, 4425], but it could equally have been due to repatterning during growth [526, 1357, 1775, 3424, 3448]. To distinguish between these alternatives, short-term cell movements (sans growth) were monitored and disctype preferences were thereby proven [1202, 1205]. A vs. P

IMAGINAL DISCS

preferences had been suspected from the behavior of en1 clones in mosaics [1358], and recent evidence supports this general notion (cf. Ch. 6) [354, 2437, 2993]. From ∼3 to 5 h AEL (St. 5–10), En states are sustained by signals that come from the adjacent Wg stripes (Fig. 4.4) [271, 1093, 2719, 3921, 4489]. Manifestations of this dependence include (1) decay of En expression in cells as they move away from the Wg/En border during segment growth [4491] and (2) premature decay of En stripes in wg null embryos [1065, 2719]. Interestingly, the dependence is mutual [1063, 1836, 1841, 4743]. Wg expression is sustained by a Hedgehog (Hh) signal emitted by the En-expressing cells [1969, 1977, 1982, 2494, 2719]. Wg and Hh are both secreted proteins [334, 574, 1789, 4440, 4455], whose modes of action are discussed later. For now, the essential point is that the diffusion ranges of Wg [271, 920, 1541, 1779, 4442] and Hh [1969, 4262] are a only few cell diameters in the embryo [3748], so any cell that moves outside this range will cease expressing whatever genes were dependent on that morphogen [2810, 2814]. The third morphogen that is instrumental in disc development is Decapentaplegic (Dpp). Cells whose states depend on morphogens can be thought of as revelers in a nightclub. If the colored lights (≈ morphogen ranges) are aimed at ﬁxed spots on the ﬂoor (≈ epithelium), then dancers can jostle (≈ divide or rearrange) into or out of a colored area, but each area will be occupied by roughly the same number of people at any given time. The system is in a steady state, despite a ﬂux of its parts. After ∼5 h AEL (St. 10) both en-expressing cells [1093, 1790, 3129] and wg-expressing cells [1894, 2013, 2534, 2536, 2683] adopt determined states that are analogous to indelible skin colors [2717]. This “Cabaret Metaphor” also pertains to how discs arise. By ∼5 h AEL (when germ band extension is completed), Wg (but not En) stripes split into ventral and dorsal pieces [176, 271, 880, 1093] with an intervening gap (Fig. 4.4). A perpendicular (A-P) stripe of Dppexpressing cells intersects the tips of the ventral (D-V) Wg stripe remnants [2003, 3833], forming a ladder of struts (Dpp) and rungs (Wg) along the trunk [827, 3122]. At each Dpp-Wg intersection in T1–T3, clusters of cells make Dll. This coincidence implied that Dll is only transcribed in cells that receive both signals [827, 833]: {Wg and Dpp}

Dll?

Along the A-P axis, Dll is activated by Wg (see below) and limited by Wg’s range of diffusion. However, dpp null embryos still express Dll [1572], so dpp cannot be needed for activation. On the contrary, Dpp must be

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

stiﬂing Dll dorsally because Dll spots elongate dorsally in dpp null embryos [1572] (cf. similar effects on salivary gland anlagen [1820]). Ventrally, Dll is conﬁned by Spitz, another secreted signal [1572, 2335]. Together, Dpp and Spitz conﬁne Dll along the D-V axis. Thus, the actual rule is {Wg and NOT-Dpp and NOT-Spitz}

Dll.

Still unexplained is how Dll can coexist with Dpp at Dpp-Wg intersections if Dpp inhibits Dll. Conceivably, some factor (Wg?) overrides Dpp’s inhibition of Dll at those points. In each thoracic segment (of wild-type embryos), the Dll clusters straddle the Wg/En border [1571]. Hence, both states (Wg and En) get incorporated into the founder population of prospective disc cells. As described above, these cells are not yet segregated into ventral (leg) vs. dorsal (wing/haltere) fates. Soon thereafter, subgroups split off (under the inﬂuence of Dpp [1572]) and migrate dorsally (tracking the Dpp stripe [1571]). Dll expression wanes in the cells as they move away from the Wg rung tips [827, 1571], so Wg is probably sustaining Dll expression (the Cabaret Scenario). Dependence of Dll on Wg was conﬁrmed by showing that Dll disappears when Wg is artiﬁcially inactivated (by heat-pulsing a t.s. mutant) at this stage [827]. The turning on of the winghaltere marker Vg in the dorsally migrating subgroups in T2 and T3 [1571, 1572] may require that Dll be turned off. That is, while Dll confers the potency to make Vg, such “competent” cells can only make Vg when they subsequently shut off Dll. This “ON-then-OFF Licensing Trick” is also used elsewhere (e.g., Dll-Hth’s launching of spineless expression in the antenna, cf. Fig. 8.3). Leg vs. wing states are imprinted at ∼6 h AEL (St. 11) by the relative doses of dorsal (Dpp) and ventral (Spitz) signals that the cells perceive [2335]. Inside the nascent discs, cells cannot be undergoing any drastic rearrangements because their blastodermal fate maps roughly match the layout of adult structures [3602, 4652]. Nevertheless, leg cells must reorganize to some extent because the future proximal cells are initially located dorsal to the future distal cells, whereas they later surround them. This situation arises because these fates are assigned by differing intensities of Dpp signal [1572]. Wing disc initiation is also more complicated (in a genetic, not a topological, sense). It involves two phases. In Phase 1, Vg (a transcriptional co-activator [1686]) is expressed (as described above) together with Escargot (Esg) and Snail (Sna) in response to extrinsic inductive cues (“{Dpp and NOT-Wg} Vg”?). Esg and Sna

91

are zinc-ﬁnger proteins whose preferences for DNA sequences are virtually identical, and so are their downstream effects (as deduced genetically) – a redundancy remarkably like the bHLH proteins Achaete and Scute [1333]. In Phase 2, Esg and Sna establish an intrinsic “wing” or “haltere” state by auto- and cross-activation (again like Ac and Sc), and Vg then comes under their control [1333]. The above scenarios do not apply to genital [679, 735, 1170, 2028, 3817] or eye discs [2933, 3041, 4146], each of which comes from several segments [2103, 4825]. Indeed, the eye disc does not acquire an A/P boundary until long after it invaginates from the ectoderm (i.e., during 2nd instar) [1635, 1637].

Cell lineage within compartments is indeterminate Nowhere is the Cabaret Metaphor more apt than within each compartment as a disc grows. Cells are free to jostle relative to the adult regions (colored areas) that they will eventually occupy (when the music stops) [2142, 3947]. This ﬂuidity was revealed by Sturtevant’s studies of gynandromorphs [4180] (cf. Ch. 1), and it was conﬁrmed at a ﬁne-grained level in all major discs by randomly marking cells at various stages of development [3441]. Because descendants of marked cells tend to stay together [3441], most cell movements must be due to passive displacements [3, 1610, 1888, 1890] (newborn cells pushing extant ones? [1525, 3515, 3518]), rather than to active migration, which would fragment the clone [532, 3422, 4671]. Clones tend to have irregular outlines [3441], and when outlines from different individuals are superimposed, they tend to overlap in virtually every region of the body [532], including the eye [189, 260, 261], antenna [3446], leg [544], wing [521, 1545], notum [521, 3007], and genitalia [1107]. The only exceptions are the compartment borders. The ﬂexibility of cell lineage is epitomized by Minute mosaics [354]. When single cells are spurred to grow faster than their neighbors (by expulsion of a retarding Minute LOF allele), their descendants can occupy more than 10 times the area that they normally would within the A or P compartment without disturbing the pattern [497, 2935]. One consequence of this plasticity is robustness [543, 2375, 2448]. Amazingly, 75% of a disc’s cells can be killed without incurring any cuticular defects [1776]. The dead cells are simply replaced by compensatory growth. Apparently, the goal of disc growth is to generate enough tissue to ﬁll all subregions, regardless of cell pedigrees: “the developing pattern seems to control proliferation, rather than the other way around” [532].

92

Strictly speaking, it is not proliferation per se that is being controlled, but rather tissue mass [4579]. Cell sizes can be altered drastically without affecting disc size, gene expression zones, or cuticular pattern [2081, 2771, 3081, 3750, 4576]. Any PI gradients that exist in discs must therefore be heedless of cell boundaries [546, 3862]. In other words, the “graininess” [1822, 2016] that cells impose on the overall “image” is irrelevant to the production of the image itself. Disc cell proliferation is sustained by circulating growth factors from the larval fat body [2707] and possibly the brain [1346]. Endocrine signals like these are transduced by the insulin receptor pathway [486, 3450] , which responds to insulin in cultured cells [919] and has significant effects on cell and organ size in vivo [2476] . Within the leg and wing discs, the chief paracrine stimulant appears to be Dpp [569, 619, 643, 1674, 1839], and the same is true for the eye disc during its early stages [570]. Nevertheless, many other pathways participate [543, 1126, 1142, 4667, 4886], including dActivin [511] , Hh [2214] , Wg [734] (although, oddly, not in the wing pouch [2254] ), EGFR [3470] (including the wing pouch [3025] and eye [185] ), JAK-STAT [1906] , and Notch [2362] . Indeed, Dpp can only elicit outgrowth of the leg [617, 4490] or wing [558, 559, 617, 643, 3862] by cooperating with an unknown factor whose expression and regulation mimic those of Wg [2254] (cf. Chs. 5 and 6). A common conduit for multiple growth-regulating pathways is the Ras-MAPK cascade [2132] , which promotes dMyc stability [1342, 2081, 3846, 4124] and blocks apoptosis [2793] . Discs also use anti-mitogenic signals that evidently rely on junctions [533, 1907, 4745] and cadherin-like adhesive proteins [1386] because LOF defects in these components can cause overgrowth [4742] . One such retardant is nitric oxide [2370]. There must not be many such restraints on growth, however, because no single or compound LOF mutants have ever been found that rival the drosophilid titans of Hawaii [669]. Indeed, no such giants have yet been produced by any GOF manipulations either [2561]. Until about mid-3rd instar, mitoses are distributed evenly in leg discs [1599], and the same is true for wing discs [20, 2015, 2848, 4427]. The uniformity is attributable to a compact (≤0.7 kb) cis-enhancer that ensures ubiquitous expression of String – a rate-limiting (Cdc25) regulator of mitosis [2480]. Around mid-3rd instar, there is a transition from random to patterned mitoses. The change is probably due to a switch from the previous (cell-size responsive?) enhancer to other (prepattern responsive?) enhancers at the string locus [2480]. Proneural areas be-

IMAGINAL DISCS

come quiescent [1925], including the future wing margin [2079, 2081, 3154, 3374]. So does a band of cells just anterior to the A/P compartment boundary [543, 3813] (possibly dpp-on cells [2739]) and a spot at the center of the leg disc [1599, 4666] (possibly aristaless-on cells [620, 2287]). In the early pupal period, most SOPs undergo their differentiative divisions, and mitoses occur in a belt around the notum, along wing veins, and then in the interveins [1545, 1741, 3813]. Proliferation ceases one day after pupariation [1312]. Cessation of growth must be controlled intrinsically because wild-type discs do not exceed their normal size when given extra time to grow (via pupariation delay [3955] or transplantation [19, 542, 546, 1357]). In theory, terminal size could be targeted in terms of total cell number [99, 480, 1548, 3483, 3492], but in fact, as stated above, growth is instead shut off at a certain mass [2078, 3081, 4576]. The unresolved “Growth Cessation Mystery” is: what local cues do cells use to monitor such a global property [3416, 4114, 4196, 4886]? One possible answer is that some morphogens also act as growth (or survival) factors [847, 3862]. In that case, organ dimensions could be limited by the range [855, 2291, 4265] and/or slope [1142, 2448, 2852] of a gradient that is steeper in young discs [977, 1169, 3213, 3862]. Indeed, discs with extra Dpp-producing zones can triple in breadth [4229, 4418]. Whatever the mechanism, it evidently operates separately in different compartments because their sizes can be manipulated independently in the same disc [1169, 3947, 3961, 4265].

The Polar Coordinate Model linked regeneration to development When mature discs are cut into pieces and forced to metamorphose, each piece forms a certain subset of elements. In this way, fate maps were charted [526, 3165]. In 1971, experiments were described where parts of leg discs were allowed to grow before metamorphosis. Regardless of whether postsurgical growth occurred in situ [522] or in an adult host [3808], the upper (stalk-end) half typically regenerated (i.e., made a whole leg), while the lower half “duplicated” (i.e., made its fated structures but in two copies). This outcome can be restated as “ABC makes ABCDEF, and DEF makes FEDDEF,” where letters are identities (∼ dorsal to ventral) and underlined elements are added postsurgically. Lateral and medial halves also obey this “Reciprocity” Rule [3808]). That is, one fragment (medial) regenerates while its complement (lateral) duplicates.

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

To explain why leg discs behave in this way, Peter Bryant proposed that cells at a cut edge can only adopt lower values in a PI-like gradient [522, 523]. If 654321 denotes the gradient (each digit = 1/6 of the disc along an axis), then this “Down the Slope” constraint means that both halves must make DEF because they share the same (∼3.5) wound edge [20]: 654 (ABC) makes 654321, and 321 (DEF) makes 123321. This “Gradient of Developmental Capacity (GDC) Model” argues that a cell’s positional value dictates its regenerative repertoire [525]. Research soon shifted to the wing disc because its larger size offered much higher resolution. Bryant bisected wing discs at 5 longitudinal and 8 transverse levels, and found that the fragment pairs obey the same Reciprocity Rule as leg discs [524, 525]. These orthogonal series suggested a Cartesian coordinate system, but diagonal cuts behaved similarly. Thus, regeneration proceeds in all directions away from a peak near the center of the disc, indicating a cone-shaped gradient (Fig. 4.5c). Tests of the model, however, produced confounding results: 1. Paradox 1 (Fig. 4.5d): the central fragments. Any piece that contains the high point should regenerate, but central pieces (bearing the gradient’s peak) actually duplicate [536]. Bryant initially guessed that loss of the peripodial membrane from these 4-cut (“cookie cutter”) fragments might be responsible for this odd result [524]. 2. Paradox 2 (Fig. 4.6a): the quadrants. The regenerative power of a fragment should only depend on its wound edge(s). However, all four quadrants duplicate, and complementary pieces regenerate, so the behavior of an edge actually seems to depend on whether it belongs to a 1/4 or 3/4 piece [526]. 3. Paradox 3 (Fig. 4.6b): intercalary regeneration. Because the gradient’s peak is central, marginal pieces must duplicate. When such pieces are cultured singly or intermixed with identical fragments this result is indeed observed. However, when pieces from opposite sides of the disc are co-cultured as a scrambled mass, they regenerate the midsection, including the peak [1773, 1775]. Intercalary regeneration had been previously documented in cockroach legs [386, 539], and the Reciprocity Rule had been found to govern amphibian limbs [20]. A concerted effort (among Peter Bryant, Susan Bryant, and Vernon French) to reconcile these diverse systems led to the “Polar Coordinate (PC) Model” in 1976 [1303].

93

In contrast to the GDC Model, the PC Model explains the Reciprocity Rule by postulating that growth depends on interactions between wound edges, rather than on the free edges themselves [354, 536]. It departs from Wolpert’s idea of morphogen gradients insofar as it relies exclusively on communication between adjacent cells [4682]. Cells are presumed to execute only two cardinal instructions: 1. “Shortest Intercalation” Rule. When cells are confronted at wound edges, they assess one another’s coordinates. Disparities provoke mitosis, and newborn cells adopt values between those of the ﬂanking cells. In so doing, the cells pursue the shorter of the two possible arcs (on a circle) that connect the points. How cells might compute or compare alternate path lengths was never clear [1805, 2291, 2808, 3862]. Because the intercalated span of values is identical for complementary fragments, only the context dictates whether the new growth is “regeneration” or “duplication,” but this distinction is moot to the cells themselves. Typically, the model is illustrated as a clock face, but cells are not supposed to perceive the “12/0” meridian as a discontinuity. Hence, the PC system is seamless, except for a singularity at the origin where opposite angular values converge like a contracted purse string. What happens to the rules at this point (the “Central Degeneracy Problem”) was never solved [3915, 3916, 4698]. 2. “Distalization” Rule. In the original version of the model, a complete circle of circumferential values was thought to be essential for newborn cells to “distalize” – i.e., adopt fates at the next level (ring) in a distal (central) direction. (How cells sense completeness was unclear.) Exceptions [3811, 4141] led to a revision [547]. Newborn cells were now supposed to distalize whenever their intended coordinates are already taken by cells at that radial level (regardless of whether a whole circle is present). The revised Rule 2 conforms with Rule 1 insofar as it also depends solely on nearest-neighbor interactions. The geometry advocated by the PC Model was not new (Driesch had toyed with polar coordinates in 1894 [1101, 3742]), but the synthesis that it achieved was historic because it deftly explained a huge mass of confusing data. Also, it was the ﬁrst real hint that arthropods and vertebrates might build their limbs in similar ways. Over the ensuing decades, the model’s popularity waxed and waned [354, 531, 863, 2420]. One lemma, at least, endures to this day: when the normal chain of events is diverted by

94

IMAGINAL DISCS

a

e map fat II

b

III IV

ions ect s i b E F

D 8 7

Key:

B

G

V

fragment duplicates

C

fragment regenerates

6A 5

I

c

4 3.5 3

AP

2

GDC Model

PA

NP

1

V

PS

A

SA

SC

4

F

P

peak

C peak

D

0

2

DC

Paradox

d

FIGURE 4.5. Regenerative potency of wing-disc fragments and an early model.

a. Fate map (abridged) of a mature right wing disc (notum, light shading; arc = suture; wing, dark shading) as per Bryant except that dots are actual SOP sites [1925] and thick lines are prevein zones (I–V) [4189]. Vein I is the bristled part of the margin [1741] (thin dashed line = bristleless part). Thick dashed line is the A/P compartment boundary (cf. Fig. 4.4). Directions (A, anterior; P, posterior; D, dorsal; V, ventral) are given in the compass at right. In all bristle pairs except scutellars (SC; bounded area = scutellum), the P partner is on the right (cf. Fig. 3.4 for abbreviations). White trapezoid is a unique spot (cf. b). b. Results of bisecting a disc lengthwise (B–F) or transversely (1–7) along folds (dashed lines) or elsewhere and letting each piece grow in isolation inside the body of a host adult [524]. As indicated in the key, an unﬁlled arrowhead means that the fragment to its rear “duplicates” (i.e., makes two copies of its fated structures as mirror images), whereas a solid arrowhead means that the fragment behind it “regenerates” (i.e., restores the whole). Note that every cut line has arrowheads of opposite type. Thus, the wing disc obeys a “Reciprocity Rule”: when one piece regenerates, the reciprocal fragment duplicates. The fact that regeneration proceeds away from D, E, 3, and 3.5 lines suggests that the disc uses x-y coordinates, but diagonal cuts (not drawn) showed that potency actually declines radially from the center piece (black trapezoid). c. The Gradient of Developmental Capacity (GDC) Model was based on the ﬁndings shown in b [524]. It proposed that the center piece contains the peak of a conical gradient of regenerative potency. The gradient is schematized here as triangles (side view) above the columnar epithelium (inscribed with imaginary contour lines). When the disc is cut, the gradient is supposed to force cells at each edge to grow down its slope. d. According to the GDC Model, any piece that includes the peak should regenerate, but the “CF24” piece actually duplicates (as does BF16; not shown) [526]. This and other paradoxes (cf. Fig. 4.6) led to the GDC model being abandoned in favor of the Polar Coordinate Model [1775]. [524],

miscues, it can produce global deformities (e.g., trefoil limbs [539, 1805]) that appear freakish to human observers but look natural to the cells themselves because no local rules are broken. The PC Model was theoretically able to encompass normal development as well as regeneration because disc growth in situ might also be regulated by intercalation [1301, 1303]. If nascent discs contain only a subset

of a mature disc’s positional value spectrum – say “1” and “5” of a ﬁnal “12345” array – then juxtaposition of 1 and 5 could lead to birth of a 3, and growth would stop with insertion of 2 and 4. PI (and pattern) would thus be elaborated steadily through such “intercellular negotiation” [834, 4682]. Growth is probably not limited in this way, however, since ﬁnal disc mass is independent of cell density [4576].

CHAPTER FOUR. ORIGIN AND GROWTH OF DISCS

a

D-E

95

Paradox

α

b

β

Paradox

8

8 6

6

3.25

cultured separately

cultured together

2

2

0

δ

11

11

9 10

10 9 9

h e als c inter

d

11 12/0

10 9

7

12/0

1 2 3

interc

4

8 7

6

heals & 9 intercalates 3 4 8 7 6 5

5

6

2

Polar Coordinate Model

alates

10

& alate s

11 12/0

10

4

8

c 12/0

11 12/0 1

11 12/0 1 10 ab 2 c d e 3 9

5

s & lates a

11 12/0 1

1

he inter al c

γ

0

2 3

9

4

8 7

6

5

10

2 3

9

4

8 7

6

5

FIGURE 4.6. Paradoxes that led to the Polar Coordinate Model (cf. Fig. 4.5 for symbol key and context).

a. According to the GDC Model, a fragment’s decision to regenerate vs. duplicate should depend only on its edges, but quadrants (cornered at the inferred peak of regenerative potency) violate this axiom. For example, the α quadrant duplicates, but a piece with the same vertical edge (3/4 complement to β) regenerates. b. Dorsal (“02”) fragments only make dorsal structures, and ventral (“68”) fragments only make ventral ones. However, when 02 and 68 pieces are scrambled together, they regenerate middle (“26”) structures. (See Fig. 4.5 for the numbering scheme.) This result is also hard to explain in terms of free edges. c. The Polar Coordinate (PC) Model avoided both these paradoxes. Based on the phenomenon of intercalation (b), this model argued that growth depends on interactions between wound edges, not on free edges. Every cell is supposed to have a circumferential (clockface) coordinate (= angle from a 12/0 reference line) and a radial (letter) coordinate (= distance from center). The “Shortest Intercalation Rule” asserts that when cells touch (via healing), they assess each other’s coordinates and ﬁll in missing values by the shorter route (no discontinuity at 12/0). For example (at right; cf. b), when “10” abuts “8,” a “9” ﬁlls the gap, instead of “11, 12/0, 1. . . 7” (the longer route). Prime marks denote new values. d. The PC Model obeys the Reciprocity Rule (see α in a). For example, when a fragment contorts to heal “9” and “12/0” edges together, this contact creates “10” and “11” (black tissue), regardless of whether the edges belong to a quadrant (left) or its 3/4 partner (right). Thus, quadrants duplicate because none has enough angular values (i.e., ≥ half ) to regenerate. The various panels are adapted from [524--526,1775].

The idea of a common cellular logic for development and regeneration was supported by (1) the ability of immature disc tissues to provoke regeneration in mature disc fragments [1183], and (2) the ability of cell-

killing agents (e.g., γ-rays [3440], X-rays [3449], UV-rays [1424], or microcautery [425]) to evoke the same kinds of mirrorimage duplications from developing discs as are obtained with fragments of mature discs. In the latter case,

96

the presumption was that necrosis-induced duplications arise via the same route as surgical tissue removal [539]. Indeed, wing-disc structures that maximally resist γ-ray-induced duplication were mapped to precisely the same spot as the GDC cone peak that was localized via surgery (Fig. 6.1) [3440]. Subsequent studies conﬁrmed some of the model’s assumptions, including the topology of wound healing [537, 947, 3564, 3565], the need for contact between cut edges [1184], and the conﬁnement of growth to a blastema at the wound interface [5, 489, 537, 946, 3155]. De novo formation of wing margin between dorsal- and ventral-type cells validated the concept of intercalation [354, 1038], and the ability of different discs to trigger intercalation inter se in a position-speciﬁc manner [18, 535, 1774, 4665] afﬁrmed the idea of a universal PI “language” [3448].

But regeneration has peculiarities that set it apart Contrary evidence, however, also accumulated. Below is a partial list of ﬁndings that deﬁed the PC scheme (see [1301, 2161] for critiques): 1. Violations of the Reciprocity Rule (e.g., fragments regenerating that should only duplicate) were found after tissue maceration [3788, 4141], during long-term (≥1 week) culture [1125, 2144, 2230], and as a function of the fragment’s proximity to – or straddling of [2139, 2143] – the A/P compartment boundary [2140, 2141]. 2. Complementary fragments should grow equally because they intercalate alike. However, duplicating fragments hardly grow at all despite adding more bristles [3808]: they neither double in mass nor grow as much as their regenerating partners [19, 20]. These heretical results challenge the dogma [525] that discs only use “epimorphosis” (patterning that re-

IMAGINAL DISCS

quires mitosis) vs. “morphallaxis” (repatterning sans growth) [2945, 4724, 4736]. 3. The eye-antenna disc should have the coordinate system of a single disc because one part (eye) regenerates and the other (antenna) duplicates [1404, 1406, 1423, 3810]. However, in other contexts (homeosis [3439, 3443, 3447], transdetermination [3788], or ubiquitous Antp expression [1416]) it behaves like a compound legand-wing disc. This “One Disc or Two? Paradox” [1805, 2933, 3788] seemed to have been settled in favor of two discs in 1997 based on the homeotic effects of Distalless [1561], but the one-disc idea rallied in 2000 based on the homeotic effects of the Iroquois Complex [697]. The rule that “antenna never regenerates” [1404, 1411, 1423, 3810] is broken when antennal tissue is macerated before culturing: in ∼20% of the implants, structures form that normally come from the eye part of the disc (ocelli, ptilinum, frons, and head bristles, but not facets) [3788]. Such regeneration should be impossible if the antenna has fewer than half the angular positional values. 4. Blastemas can form at both edges of a fragment [2213] and can regenerate structures in a correct circumferential sequence (as per the fate map) [2144] before healing together. This behavior is more consistent with the GDC Model [523] than with the PC Model. One challenge that the PC Model initially seemed to deﬂect was the sequestering of regenerative potential in a single quadrant of the leg disc [3808]. As explained in Ch. 5, the model attributed this anomaly to a clustering of angular values [1303, 4140], but a recent molecular analysis of wound healing [1472] reveals trans-lumenal interactions that have mortally wounded the PC Model itself.

CHAPTER FIVE

The Leg Disc

In an article entitled “Pattern formation in the embryo and imaginal discs of Drosophila: What are the links?” [4675], Adam Wilkins and David Gubb posed a question that was on the minds of many researchers at that time (1991). The embryo’s segmentation hierarchy was basically understood, but it was unclear what these various genes might be doing in discs [4682]. Wilkins and Gubb argued that segment-polarity genes supply the angular values of the Polar Coordinate (PC) Model, which until then had only been an abstract formalism. Several predictions of this “Angular Values Conjecture” were soon put to the test by molecular genetics. Chief among them was the expectation that LOF and GOF alterations in segment-polarity genes should reorganize disc anatomy. This prophecy was indeed fulﬁlled, and the experimental probing uncovered a trove of insights into the machinery of disc patterning. The conjecture itself, however, turned out to be wrong. Segment-polarity genes do not paint a pinwheel on each disc. Rather, they draw a few important lines – the compartment boundaries.

The Molecular Epoch of disc research was launched in 1991 The Wilkins-Gubb paper, with its clarion call for a molecular assault on discs, provides a convenient demarcation between the Cellular and Molecular Epochs of disc research. In the 1990s, many links were patiently forged between the blastoderm and the adult. During this process, several old puzzles at the cellular level were solved by clever experiments at the molecular level. Among those puzzles were the following [1807]:

1. “Gaps Mystery” (Fig. 5.1e). Cellular Problem: Each leg’s bristle pattern is roughly symmetric about the D-V plane [1883], as are afferent projections of the bristle neurons [3005]. It therefore came as a surprise when the A/P compartment boundary was found not to coincide with this plane, but rather to be offset from it by several cell diameters [1800, 2449]. Shouldn’t they be congruent if this line is used as a reference axis for specifying mirror-symmetric positional information [904]? Molecular solution: The reference axis turns out not to be the A/P boundary itself but a morphogen-producing zone just anterior to it [1807]. There are no gaps between that zone and the mirror plane. 2. “Quadrant Regeneration Mystery” (Fig. 5.2a). Cellular Problem: In contrast to the wing disc, where every quadrant duplicates, one quadrant of the leg disc regenerates [3808, 4140]. Why is this upper medial (UM) quadrant special? Molecular solution: Its peripodial membrane turns out to have a spot of gene expression that dictates regeneration vs. duplication [1472]. The dogma that regeneration and development always use the same rules is false. 3. “Dorsal Remnant Mystery” (Fig. 5.2b). Cellular Problem: When embryos are exposed to X-rays [3449] or UV-rays [1424] at doses that are sufﬁcient to kill cells [2774], the legs that manage to complete development often manifest a “duplication-deﬁcient” anatomy, where part is missing and the rest is duplicated in mirror image. Similar defects can be produced by microcautery [425], by heating t.s. cell-lethal mutants [101, 2102, 3702, 3964], and by in situ excisions [522]. First legs tend to fuse rather than duplicate [3442] because their

97

98

discs are conjoined [377, 3426], but regardless of how the symmetry arises, the remnants tend to retain dorsal structures at the expense of ventral ones [425, 2102, 3442, 3449, 3705]. Why is the dorsal region (outside the UM quadrant) more robust? Molecular solution: It is the source of a diffusible growth factor that stimulates mitoses throughout the disc [529, 1807], and this growth factor is upregulated in response to trauma [2856]. 4. “Triplications Mystery” (Fig. 5.2c). Cellular Problem: Triplications can also be induced in cell-lethal

IMAGINAL DISCS

mutants during certain sensitive periods [1495, 2102]. Typically, such legs have their normal tip (stem) plus an outgrowth (branch) that either diverges or converges (i.e., gains or loses circumferential elements distally). Outgrowths that branch from the stem’s ventral face tend to diverge, while dorsal ones tend to converge [1495] – a bizarre trend also documented in cockroach legs [384, 547, 2804]. Why? Molecular solution: The deciding factor appears to be the mixture of morphogens on one side of the disc vs. the other that is created under these conditions [2856].

CHAPTER FIVE. THE LEG DISC

99

FIGURE 5.1. Geometry of leg development.

a. Mature, left 2nd-leg disc (∼150 µm across [3140]). Lines are folds. The central circle is the “endknob.” Connections to larval skin (stalk) and CNS (nerve) have been cut. b. Longitudinal section (“BM” and “PM” = basement and peripodial membranes; “CE” = columnar epithelium). Transverse lines are cell boundaries. The black cell is one of several larval neurons that innervate Keilin’s organ [834, 2287]. These neurons invaginate with the disc (a piggybacking that is ancient in dipterans [2824]). Their dendrites (not shown) stay attached to larval skin through the stalk [4330], while their axons project to CNS (a route later taken by adult axons). Adepithelial cells are omitted. c. Idealized fate map of left 2nd-leg disc. Segment primordia are the coxa (Co), trochanter (Tro), femur (Fem), tibia (Tib), basitarsus (Ba), and tarsal segments 2–5 (unmarked), with claws mapping centrally. The sternopleura (Stpl) is part of the body wall (cf. Fig. 4.4). AC and PC (shaded) are A and P compartments. Spokes denote bristle rows, whose spacing is estimated: it may not be so uniform (cf. e). As shown below for the tarsus, the annuli telescope out during evagination (toward the viewer) as the covering PM (not shown; cf. b) peels away [1311]. In this schematic, the tarsus also rolls clockwise ∼90◦ . Compass directions (D, V, A, P) refer to prospective or actual adult axes (cf. d, e). d. Anterior face of left 2nd leg (∼2 mm long). There are 2 macrochaetes (distal tibia) and numerous chemosensory (curved) bristles on the tibia and tarsus. “EB” (edge bristle) is at the D midline. Tarsal rows 5–8 are marked on the compass below. e. Basitarsus (whole surface) drawn as if slit along D midline and spread ﬂat. Most bristles have a bract (triangle) and align in rows (numbers at top). Five chemosensory bristles lack bracts and reside between rows, as do 3 sensilla (circles at 3.5, 5.5, and 8.5). Between rows 1 and 8 is a lawn of hairs. Row-1 and -8 bristles are peg shaped and darker; others are tapered and lighter. Aside from bristle thickness and pigmentation, other useful landmarks include bristle lengths and intervals – both of which increase linearly from V to D (graph at top). Indeed, the symmetry is so precise [1883] that it was surprising when the A/P compartments were found to be offset from this D-V mirror plane by a few cells (“gaps” below) [1800, 2449]. Row-1 bristles can reside in either AC or PC (cf. Fig. 3.9c). Panels a and b were traced from [377] and [3426], respectively (see [1774, 2144] for nerve-stalk asymmetry, [1516, 2287] for folding details, and [1311] for region-speciﬁc cell shapes); c is adapted from [1807, 3807]; and e is based on data in [1801, 1803]. N.B.: For reasons explained in Fig. 4.4h legend, some authors orient the axes (c) with the D pole pointing into the stalk (at 12 o’clock vs. ∼1 o’clock here). Claws (c) actually map more dorsally [1812, 3807, 4140]. Numbering of rows (e) follows original nomenclature [1714]. The numbers are backward in [837, 1039, 2533, 4159].

Before examining how these and other riddles were demystiﬁed, it may help to review some of the clues that accumulated before the Molecular Epoch. Those clues tantalized two generations of researchers, and the older analyses were at least as interesting as the studies of the recent past.

Bateson’s Rule (1894) governs symmetry planes in branched legs The latter three mysteries listed above are interrelated insofar as they all involve mirror-symmetric partial patterns. Near the end of the nineteenth century, William Bateson collected numerous examples of such abnormalities as naturally occurring “sports” in various species of arthropods [240]. In virtually every 3-tipped leg, he found that (1) the three tarsi lie in one plane (vs. a tripod), and (2) the outer two are mirror images of the inner one [2809]. These constraints constitute “Bateson’s Rule.” Although the tips always manifest R-L-R or L-R-L symmetry (R and L = right vs. left-handed), the inner member can face the ﬂanking ones with any surface (D, V, A, or P). To illustrate this rule, Bateson built a cute wooden model (Fig. 153 in his 1894 tome) where the bases of three tarsi are geared together. When the in-

ner one rotates clockwise, the outer ones turn counterclockwise, thus maintaining mirror symmetry. Bateson’s Rule must reﬂect a deep law of cellular “psychology” because the syndrome that it describes transcends the mode of tissue removal (irradiation, surgery, or apoptosis) [547]. The PC Model invoked a cellular imperative to “Always maintain continuity!”: when distant cells are apposed, they multiply to ﬁll the gap in positional values by the shortest possible route [539, 2513]. With this basic edict and one ad hoc assumption, the PC model could explain why the UM quadrant regenerates (Fig. 5.3b). To wit, this quarter might contain >50% of the disc’s angular coordinates [1303, 4140]. However, the idea of crowded coordinates undermined the model’s other proposal that discs stop growing when a certain density of coordinates is reached (like people ﬁlling seats in a stadium) [546, 1303]. The PC Model was also able to account for duplications, triplications, and other outgrowths [538, 1496]. Those conjectures were supported by clonal analyses [1497, 1503, 1504, 2102], apoptosis studies [797, 1501], and phenocopy simulations [1498]. In short, the model was robust. Even so, there were doubts. Some of arguments that were contrived to defend the model strained credulity

100

IMAGINAL DISCS

FIGURE 5.2. Mysteries of leg development (cf. text and Fig. 5.1e).

a. Quadrant Regeneration Mystery. Unlike the wing disc, where every quadrant duplicates, the upper medial (UM) quadrant of the 1st leg disc regenerates (black triangles; cf. key in Fig. 4.5). UM regeneration is seen either with or without the endknob (center region) [3808, 4140]. b. Dorsal Remnant Mystery. When cells are killed by X-rays [3449] or cell-lethal mutations [2102, 3705], D structures tend to remain, whereas V ones are lost (except near branch tips). First-leg discs, which are conjoined [3426], fuse medially instead of duplicating, and losses are symmetrical on left (L) vs. right (R). In this contour map, darker zones denote higher frequencies (0–30; –60; –90; –100%) of elements present in 65 pairs of fused 1st legs. c. Triplications Mystery. Triplicated legs possess a “stem” [1875] and a symmetric outgrowth. V-side outgrowths tend to diverge, whereas D-side ones converge, and the (2/6.5) line between cohorts is sharp. Maps in a, b, and c are redrawn from [3808], [3442], and [1495], respectively. N.B.: “Medial” and “lateral” mean toward or away from the larval midline, whereas A, P, D, and V refer to prospective or actual directions in the adult.

because they demanded odd regions of cell loss and contorted angles of healing (Fig. 5.3c–e) [1495, 3442]. Moreover, much of the later data came from a cell-lethal mutation [1502, 3945] whose gene product was then unknown. (Suppressor of forked turned out to be a polyadenylation protein that affects mRNA stability [2879, 4235] and licenses metaphase [140, 141].) Skeptics felt that the model relied too heavily on ad hoc fantasies and molecular phantoms [863, 2161].

Meinhardt’s Boundary Model deftly explained Bateson’s Rule In 1980 (four years after the PC Model’s debut), Hans Meinhardt, a mathematically gifted theoretician, offered a clever new solution to Bateson’s Rule (Fig. 5.3) [2804]. His “Boundary Model” was based on the leg’s A/P compartment boundary and a supposed D/V bound-

ary that subdivides the A region into D and V parts [4076]. Meinhardt inferred that distal tips grow out wherever P, V, and D areas convene [2807]. Normally, this condition is only met at the distal tip, but extra tip “organizers” could arise if P, V, or D cells were to appear in the wrong place due to wounding or apoptosis. The idea that trauma could trigger ectopic gene expression seemed farfetched at the time, but it has since been conﬁrmed [489, 1472, 2856]. As for how three territories might cooperate to induce distalization, Meinhardt conjectured that each cell type secretes an ingredient needed for synthesis of a tip morphogen “T.” That is, the P, V, and D molecules react chemically to make T. Each compartment may be responsible for a particular step in the synthesis of the morphogen or . . . may produce a diffusible cofactor which is required for morphogen production. [2804]

CHAPTER FIVE. THE LEG DISC

The ﬁctional P, V, and D molecules should not diffuse far beyond their source sectors since, in that case, tips would form everywhere. The model thus invokes three short-range inducers and one long-range morphogen. T forms a cone-shaped gradient that could encode radial distance like the PC system’s radial coordinate [2805, 2813]. In Meinhardt’s original model (1980), there is no cellular imperative about continuity. Rather, Bateson’s Rule is an emergent property of the model’s criterion for distal outgrowth [617, 2706]. The geometry is easy to visualize (Fig. 5.3g–i) [2808, 2809] but hard to describe in words. Imagine that the random switching of cell states is simulated by spattering P-, V-, or D-colored “paint” onto a tricolored pie chart. Wherever the “P + V + D = T” criterion is satisﬁed, a tip will grow out. Among the possible conﬁgurations are 1. A dab lands wholly inside a sector. In this case, nothing happens because there would only be a bipartite reaction (P + V, V + D, or D + P) at the spot’s edge. 2. A dab lands at the perimeter and straddles a boundary (Fig. 5.3g). If the spot’s color matches a ﬂanking color, then nothing happens, but if it differs from both, then T will be made where the three colors meet, and an extra tip will emerge in mirror image to the endogenous tip (for reasons described in 3). 3. A dab straddles a boundary but does not reach the perimeter (Fig. 5.3h). If the spot’s color differs from both ﬂanking colors, then T will be made wherever its edge crosses the border. Rounded spots will have two T points, and the line connecting them must intersect the disc’s center because they lie on a radius (P/V, V/D, or D/P). Thus, the ﬁrst part of Bateson’s Rule is satisﬁed: the three tips will lie in one plane. Because each T point has a pinwheel (PVD) handedness, a person in the P zone near a juncture must walk either clockwise (L) or counterclockwise (R) to step into V and then into D. Along a radius crossed by a dab, the handedness changes at every T point, thus fulﬁlling the second part of Bateson’s Rule (i.e., R-L-R or L-R-L alternation). 4. A dab’s edge touches, but does not cross, a boundary (Fig. 5.3i). T will be produced on either side of the intersection, and a pair of tips begins to grow out but eventually converges (vs. diverges as in 2) because (1) “the distance between the two [tips decides] . . . how many elements are missing . . . due to overlap of the two gradients,” or (2) “the normal maximum concentration is not achieved [because] . . . too few

101

cells of a particular compartmental speciﬁcation are available or if they are not close enough” [2804]. The leg’s bristle pattern is too rich to be encoded by three broad angular values (P, V, D sectors) [3253]. Meinhardt recognized this problem and in a 1983 paper sketched a “Hybrid PC-Boundary Model” (Fig. 5.3k) where the P/V, V/D, and D/P lines provide three angular values in the nascent disc [2808], with remaining values emerging later via the Shortest Intercalation Rule. One difﬁculty with this remedy is that all three spokes lie outside the coordinate-dense UM quarter, so it should be unable to regenerate. In the same article, Meinhardt concocted an alternative solution, which will here be called “Boundary Model II” (Fig. 5.4). Its key amendment was that “the compartment borders must act as a frame for the ﬁner circumferential subdivision, [with the] distance from a particular compartment border [being] measured by a diffusible morphogen” – an idea earlier suggested by Francis Crick and Peter Lawrence [904]. Echelon 1 (at the boundaries) P+V=A V+D=B D+P=C Echelon 2 (at the center of the disc) A+B+C=T T would thus be created in two steps (vs. one in Boundary Model I), and its source point would be deﬁned by the intersection of three lines (vs. the juncture of three areas). More important, an angular coordinate could be cobbled from the new factors. A, B, and C are supposed to be long-range morphogens (like T but unlike inducers P, V, and D), so they could form tentshaped gradients [2804]. Cells could then ﬁgure out where they are by measuring concentrations of A, B, and C. To encode angular PI, each gradient must curve around the circumference (Fig. 5.4a), a constraint that a priori seems implausible [618, 2456]. Conceivably, the molecules could be channeled or actively transported (see below). Cartesian and polar systems could easily be interconverted [922, 1298, 2157, 3252, 3703], but their coordinates cannot be swapped directly inter se [548, 2512, 2881, 3915, 4697]. Boundary Model II also explains why some disc fragments duplicate (Fig. 5.4b). Thus, for example, if the D sector is excised, then the P and V edges should synthesize A (when they heal together) at a site opposite to the extant A source [2808]. If the new A gradient respeciﬁes cells by morphallaxis, then it will cause a mirror-image

102

IMAGINAL DISCS

a 8 910

11 12/0

b678

1

7 6 5

e dcb

a

1

4

L

R

L

L

1

1

L

2 4

1

b

7 6 5

L

2 4

3

2

L

3

k D V

R L

P

P

P

explains:

D

88%

... but not:

D

D

D

P

V

P

25% V

V

0%V

P

D

D

16% P

D P

19% V

L

V

D

L R

V

D

D

V P

a e dcb 2 P 3

hybrid

h i

j

V 4

P

V

D

6 5

g

conical gradient

L R L

9 10 11 12/01

7

P V

10 1112/0 11 10 11 9 11 L 12/0 8 12/0 1 12/0 a 7 bR 1 1 6 1 2 5 2

8

Boundary Model D

3

4

3

Polar Coordinate Model

f

2

4

3

d e8 910 1112/0a

10 11 12/0 9 8 7 6 a 5 b

1

L

2

4

3

R

1

2

3

10 11 12/0 9 8 7 7 6 b5 9 8 8 10 4 7 11 3 12/0 6 5 1 a 2 4 5 4 3

11 12/0

7 6 5

5

2 4

c 8 910

w

12/0 2 1 11 3 11 12/0 4 10 9 5 6 8 7 6 5

ne

10 11 12/0 12/0 9

V 80% P

Paradox

CHAPTER FIVE. THE LEG DISC

duplication. In accord with this scenario (but not with the epimorphosis required by the PC Model), duplicating fragments grow negligibly and lose structures near wound edges [20].

The Boundary and PC Models jousted in a ‘‘Paradigm War’’ Given the reasoning in the original Boundary Model, distal outgrowths should only sprout from dorsal, ventral, or anteroventral faces of the leg (P/D, P/V, and D/V radii). In this regard, it surpassed the PC Model (where constraints are looser) at explaining why outgrowths tend to lie in the D-V plane [617, 4490]. It also afforded a simpler view of how a disc’s coordinate system might be “geared” to the segmentation gene machine of the embryo [2812, 2813]. To wit, the A/P boundary in each thoracic segment would be recruited as a reference axis for each leg or wing disc [2808] – a conjecture veriﬁed by

103

later molecular studies (Fig. 4.4) [834]. Ironically, Boundary Model II was stumped by the same Quadrant Regeneration Mystery that the PC Model handled relatively easily (Fig. 5.3j) [2804]: how could D-type cells alone be able to regenerate a tip? Despite their ability to “cross-hybridize” (Fig. 5.3k) [1807, 2814, 3253], no two models ever differed so fundamentally as the Boundary and PC Models [834]. Their rivalry centered on several pivotal issues [354, 1807]. 1. Who’s talking and who’s listening? According to the PC Model, every cell identiﬁes itself to its neighbors and receives such input in return. In the Boundary Model, only a few cells emit signals; most are silent receivers. 2. Do cells shout or whisper? For the PC system to work, cells need only converse with their neighbors, and they are thought to do so by direct contact [1301],

FIGURE 5.3. Models for pattern formation that were prominent in the 1980s, and how they explained regeneration-duplication

phenomena in leg discs (cf. Fig. 5.2). In the Polar Coordinate (PC) Model (a–e) [1303], growth is supposedly stimulated when distant cells abut via healing, whereas in the Boundary Model (f–j) [2804, 2808] distal growth is provoked when three kinds of cells convene. a. Polar coordinates in a left leg disc (cf. Fig. 4.6). Each cell assesses its position as (1) radial distance (lettered rings) from the center and (2) angular declination (numbered spokes) from an arbitrary radius (12/0). Unlike the wing disc where angular values are presumed to be spaced evenly, here they must be crowded because the UM quadrant regenerates. b. In the PC Model, fragments (e.g., UM) with >50% of the angular values regenerate (black triangles), whereas pieces (e.g., 3/4 part) with <50% duplicate (hollow triangles). In each case, the edges (5 and 12/0) intercalate via the shortest route (cf. Fig. 4.6d). c–e. Distal outgrowths can occur if tissue is lost from certain regions, and the edges of the hole (white “sausages”) heal as shown. c. If cells are lost from the UM perimeter, then an extra “clockface” arises when new tissue (black) ﬁlls the 5–12/0 gap by the shorter route (viz. 1–4). Because the new clockface is backward, the outgrowth will be right handed (R vs. L). Comparable losses from other peripheral parts should not form whole circles, which is problematic because mirror planes of twinned legs tend to be at right angles to spoke 4, not 8 [3442]. d. Loss of internal coordinates can generate two whole new circles if wound edges reorient (during evagination?) to abut values from ≥50% declination (e.g., 9 and 3; “a” and “b” = outer two rings) [1495]. e. Confrontations of <50% declination will form partial circles and stunted outgrowths. L-R-L handedness (d, e) obeys Bateson’s Rule (see text), but the model does not explain why converging outgrowths (e) occur near angular value 1 (cf. Fig. 5.2c) and not near 2–4. f. In Meinhardt’s Boundary Model, the anterior compartment is subdivided into ventral (V) and dorsal (D) parts, whereas the posterior (P) compartment is not subdivided. The PVD sequence is clockwise in left legs (curved arrow). Each cell identity – P, V, or D – furnishes one key ingredient for making a morphogen. The morphogen is created only at the nexus, and its diffusion forms a conical gradient (rings = contours; epithelium in side view). g–i. The model explains outgrowths. g. Duplications arise when a peripheral patch that straddles a boundary (e.g., D/V) transforms into the third type (P), thus creating a new PVD nexus (arrow) that grows out to form a right limb. h. If such a patch is internal, then limbs arise at two new PVD points. i. If the patch merely touches a border, then too little morphogen may be made to sustain two tips, in which case the outgrowth converges. As for why ventral outgrowths diverge and dorsal ones converge (cf. Fig. 5.2c), the model must assume (ad hoc) that D spots tend to straddle P/V (h), whereas V spots only touch D/P (i). j. The model (in a leg disc at left) explains most surgical outcomes. When the boxed fragments are cultured, they regenerate claws (bicurved symbol) and other distal structures that map in the excised endknob at frequencies indicated. Regenerative ability is maximal when all 3 domains remain (88%), as expected. However, the model does not explain why the UM quadrant (sans endknob) also replaces claws often [2804]. k. Hybrid PC-Boundary Model, where P/V, V/D, and D/P radii specify angular values (4, 5, 1) in a PC system [2808]. Panels a and b are redrawn from [1303] (angular spacing was later revised [4140]); c–e are adapted from [1495, 1496, 3442]; f–k are based on [2804, 2808] (he uses AD and AV, not D and V); data in j are from [3811, 4140]. See also App. 7.

IMAGINAL DISCS

D

gradie ved nt r u C

b

s

a

P+V=A V+D=B D+P=C

c

104

A+B+C=T

B V

A' V

A conical gradient

Boundary Model II

c dAC

b

gradie ved nt ur

Hh + vAC = Wg Hh + dAC = Dpp

L

Dpp

Wg + Dpp = T T

PC Hh

P

A

(T>5) = Dll

Hh

vAC

V

P

A

s

c

P

T

P

Hh-Dpp-Wg Model

V A

Dll

Wg Arc Scenario

?

n

ra

Cloud Scenario om diffusio nd Dpp

?

Wg FIGURE 5.4. Molecular validation of the basic premises in Meinhardt’s Boundary Model.

a. The Boundary Model II (as revised by Meinhardt in 1983 [2808]). In this version (cf. Fig. 5.3), there are 3 new morphogens (A, B, C). Each of them arises at an interface (P/V, V/D, D/P) where region-speciﬁc molecules (P, V, D) interact after diffusing a short distance into the adjacent region. A, B, and C then react at the center to form the tip morphogen T. Finally, T diffuses to form a conical gradient that provides the radial coordinate of a polar system, while the angular coordinate is speciﬁed by the graded concentrations of A, B, and C. The gradients must overlap (not shown) to encode every angle uniquely. b. According to this model, a VP fragment should duplicate because V and P will create a second source of A (A ) when they heal together. c. Hh-Dpp-Wg Model. Developing legs actually use a strategy like Meinhardt’s model, although the ﬁrst induction involves one-way (not bidirectional) diffusion (arrowheads), and there are only 2 (not 3) long-range morphogens. Cells in the Posterior Compartment (PC) make Hedgehog (Hh) – a short-range signal. When Hh diffuses across the A/P border, cells in the ventral AC (vAC) respond by making Wingless (Wg), while dorsal (dAC) cells make Decapentaplegic (Dpp). How Wg and Dpp travel is not known. One idea (Arc Scenario, above) is that they move in arcs to form curved gradients, with subsequent steps (e.g., speciﬁcation of angular values) as in Boundary Model II [1807]. The gene Distal-less (Dll) is shown being activated above a threshold (5) in the T gradient (assumed to be 10-to-1 centrifugally). Another idea (Cloud Scenario, below) is that Wg and Dpp diffuse randomly [618, 2456], in which case they could activate downstream genes directly (rightmost arrow) where their clouds overlap (via the rule “Wg + Dpp = Dll”) without any intermediate tip morphogen [2813]. Indeed, one Wg target gene (Dfz3) does appear to be expressed in a parabola-shaped domain [3977]. However, defect arcs in genetic mosaics [1811] are more easily explained by the former scheme (see text). The Arc Scenario may also apply to the developing notum (cf. Fig. 6.14), where the dpp-on stripe is at a ∼30◦ angle relative to where its target gene (wg) is turned on [4369]. Panels a and b are based on [2808] (p. 381, para. 2), and c is adapted from [618, 1807, 1991, 2456].

CHAPTER FIVE. THE LEG DISC

whereas in Meinhardt’s scheme, cells must broadcast signals over distances of tens or hundreds of cells using diffusible molecules as messengers. 3. Is the coordinate system seamless? This issue has been debated ever since compartments were discovered in 1973 [2161, 2513]. In the PC Model, the circumference has no discontinuities. In Boundary Model II, compartment boundaries are distinctive as reference axes (as in a Cartesian system) [2808]. As the Cellular Epoch drew to a close in the late 1980s, evidence was mounting for the Boundary Model. Peculiarities of the A/P border (e.g., the two below) were difﬁcult to reconcile with the PC Model’s notion that all angular values are equivalent. The leg disc offered the ideal battleﬁeld because its geometry ﬁts a polar coordinate system so neatly (leg segments = rings, and bristle rows = spokes; Fig. 5.1c) [1311, 1807], and its A/P line is as inviolate as the wing’s [4076, 4254]. Nevertheless, the wing disc was better known in terms of growth and lineage, so some of the skirmishes took place there instead. Certain properties of the A/P border were problematic for the PC Model. For example, regenerative ability was found to be correlated with the cut’s location relative to the A/P line. Compartments had originally seemed irrelevant for PI because they can be reestablished during growth of wing- and leg-disc fragments [5, 354, 2213, 4223] or during apoptosis-induced duplication of legs in situ [1504, 3701], and Bryant’s map of regenerative potency revealed no discontinuity at the A/P line (Fig. 4.5). However, when Jane Karlsson reinvestigated using other cuts, she found a clustering of angular values near the wing’s A/P boundary [2140]. No such correlation is seen in the Drosophila leg disc [4140], but circumferential intercalation in the legs of other (hemimetabolous) insects is constrained by A/P lineage boundaries [1297, 1299, 1301, 1302]. The PC Model was also unable to explain why the A/P border region is a haven for slower growing cells in the wing blade. In the developing wing (although perhaps not elsewhere in that disc [1431, 2935, 4703]), cells that divide more slowly than their neighbors are winnowed from the population [2229, 2935, 3949, 3961, 3962], except near the A/P and D/V compartment borders [3947, 3948]. “It is as if cells at a compartment boundary form a competitive pool that is separate from that of the internal cells” [1639]. In the case of the leg disc, the gynandromorph fate map is compressed on either side of the A/P boundary [1800] – indicating less mixing of unrelated cells

105

there – but this trend might not involve competition per se [3961]. At the dawn of the Molecular Epoch, many genes were found to be expressed in sectors (Fig. 5.8) or rings (Fig. 5.11) within the leg disc. These patterns were hailed as conﬁrmation of the PC model [531, 880, 1304, 1574, 1970], but subsequent analyses of the gene circuitry tipped the scale in favor of Meinhardt’s model [617, 2814]. The awkward truth, however, is that much of the new data does not ﬁt neatly into either scheme. Chief among these “ugly facts” are the mirror-image duplications manifest by wg LOF and dpp LOF legs (Fig. 5.5), which imply a bipolar coordinate system [1812].

Hh, Dpp, and Wg are the chief intercellular signaling molecules Hh, Dpp, and Wg are all secreted signaling proteins. Hh and Wg are founding members of the Hedgehog [1701] and “Wnt” (Wingless and Int-1 [3148]) families [3923], while Dpp is a member of the TGF-β family [3099]. During the 1990s the molecular players in each transduction chain were ﬁtted together, like clues in a Sherlock Holmes mystery, to reveal the routes from the ligands to the nucleus. Among the surprises in these pathways were 1995

1996

1997

1998

1999

Discovery (in several labs) that Hh transduction involves a cAMP-dependent kinase (DC0 encodes its catalytic subunit) whose only known developmental role until then was in slime molds [355, 2115, 3343]. Identiﬁcation of the long-sought Wg receptor (Dfz2) [310, 1971, 3656] and Hh co-receptor (Smo) [61, 4441], which were unexpectedly alike [3203, 3344, 4708]. Identiﬁcation of Pan as the ﬁnal link in the Wg chain [519, 1088, 4439], which solved one mystery but led to a deeper one: why should the pan (Wg pathway) and ci (Hh pathway) genes be adjacent (∼10 kb apart)? Realization that the Hh and Wg pathways also share an unsuspected Slimb-dependent proteolysis of their transcription factors (Ci and Arm) [2060, 4279]. Revelation that the proteoglycan Dally is a co-receptor for both Wg [641, 2556, 3309, 4400] and Dpp [2004, 4400] – hence challenging the idea that Wg or Dpp are freely diffusible, as had earlier been disputed for Hh [277, 4275].

Figure 5.6 sketches the Hh, Dpp, and Wg transduction pathways. The players within each pathway are

106

IMAGINAL DISCS

a

b PAB

c wg

LOF

d dpp

LOF

D

6 7 8

5

W g1

2

V

V

Dpp 4 3 2

7 6

6

5

5

4

8

4 3

3 2

D/D

1

7

2

8 1

V/V

6

5

4

7

D

V/-

5

8

3 1

2

dpp

Cl

LOF

Deficiency & duplication Deficiency only FIGURE 5.5. Phenotypes of wg LOF and dpp LOF legs that argue for a bipolar coordinate system instead of Meinhardt’s 3-azimuth model. For gene abbreviations, see App. 6. a. Anterior face of a wild-type left 1st leg from midtibia to the distal tip. Ventral (V) and dorsal (D) sides face left and right. Landmarks include pre-apical bristle (PAB), sex comb, and claws. The apical bristle (not labeled) is less prominent on 1st (vs. 2nd) legs. At right is a fate map (cf. Fig. 5.1c). Inner circle is the tarsus (1–8 are bristle rows). The sex comb begins as a ventral transverse row; claws and PAB are dorsal [1812]. Crooked white line delimits Dpp (∼225◦ sector) and Wg (∼135◦ ) “domains of inﬂuence” (bounded by rows 2 and 7; cf. Fig. 5.2c), based on LOF defects (b–d). Because these sectors are so much larger than the slivers where dpp and wg are transcribed (cf. Fig. 5.4c), Dpp and Wg must act as secreted signals [834], but whether they are bona ﬁde morphogens is debatable. These sectors may actually be more equal in size because D bristle rows are closer to one another than are V rows (cf. Fig. 5.1e). b. In wg LOF legs, V structures are missing and replaced (cf. key at bottom) by a set of D structures arranged in mirror symmetry [880, 1812]. This D/D (“Janus”) symmetry extends through the sternopleura [177]. D/D legs also arise in arm LOF [3317], dsh LOF [2262], gam LOF [599], pan LOF [519], porc LOF [2110], and sgl LOF [1673] ﬂies and are inducible by wg null [1811], arm null [3317], and probably arr null [4570] V clones (cf. Fig. 5.6). D/D legs can also be created by heat-pulsing t.s. NLOF larvae [886] – a perplexing result because there is no known role for the Notch pathway in Wg signaling in legs. The few sex comb bristles that remain fail to rotate. A milder D/D duplication is seen in Egfr LOF gro LOF double mutants [3465]. Finally, various kinds of treatments that cause trauma (microcautery, irradiation, etc.) can produce D/D phenotypes [425, 2102, 3442, 3449, 3705], presumably because the dpp-on D sector has more robust growth potency than the wg-on V sector (cf. Dorsal Remnant Mystery, Fig. 5.2b) [1807]. c, d. In dpp LOF legs, D structures tend to be missing from the tibia and tarsus [1812]. Often they are replaced with V structures (c), but sometimes not (d). c. V/V phenotype. When a second set of V structures arises, the ﬁrst legs have U-shaped sex combs. Janus V/V legs are also seen in punt LOF [3932, 4277] and tkv LOF ﬂies [753] and are inducible by dpp null [1811] or punt LOF D clones [3329]. d. “V/-” phenotype. Missing structures in dpp LOF legs are commonly not replaced. (No analogous “D/-” phenotypes are seen with wg LOF legs.) In such cases, the V remnant tends to shorten and curl dorsally. These ﬁgures are based on [177, 1811, 1812, 3317].

CHAPTER FIVE. THE LEG DISC

outlined in App. 6. Some familiarity with these molecular components is essential to grasp the tissue-level narrative that follows.

P-type cells use Hh to ‘‘talk’’ to A-type cells nearby Based on detailed analyses of the above pathways, we now know the raison d’ˆetre for A and P compartments in most discs: they establish separate groups of “signaler” and “receiver” cells [354, 2448, 4488, 4848]. The P group emits a signal (Hh) to which the A group can respond [571, 1078, 4229]. Limited diffusion of the Hh signal ensures that only a subset of A cells (those at the border) will actually respond [358]. This “Short-range Inducer Model” thus relies on a principle that henceforth will be called the “Deaf-Speakers/Mute-Listeners Trick.” A vs. P identities in thoracic discs are implemented by two different transcription factors: En (homeodomain class) [1031, 3719, 3860] and Ci (zinc ﬁnger class) [65, 155, 1827, 3194, 4503]. In P cells, En somehow activates hh transcription [1647, 1659, 4227, 4848] so P cells secrete Hh. Simultaneously, En represses ci transcription [1135, 1647, 2581, 3818] so P cells are refractory to their own signal (because Ci is needed for Hh transduction). All A cells can detect the signal because they express every transduction component (Ptc, Ci, etc.), but most A cells should be unable to emit Hh because Ci-75 represses hh [155, 1078, 2832]. The asymmetric roles of A vs. P cells in this discourse [354, 2448, 4848] contrast with the mutual speaker–listener roles envisioned by Boundary Model II [2808], but the two models are similar beyond this point [1427, 3497]. When secreted Hh spreads across the A/P line [1472, 4228], it rescues the activator form of Ci (Ci-155) from the proteasome guillotine so Ci-155 can convey its signal to target genes. In summary, three distinct zones are maintained along the A-P axis of each thoracic disc as a result of this circuitry (Fig. 5.7): 1. P-type cells express En, which activates hh but suppresses ci and dpp [4229]. En’s inhibition of dpp occurs by two routes: En directly binds dpp’s cis-enhancers [2980, 3747], and En blocks ci expression so Ci-155 cannot activate dpp [1647, 1827, 2832]. 2. A-type cells at the border receive Hh signal that diffuses from the P side. They respond (via PKA and other agents) by rescuing Ci-155 from being cleaved so it can activate dpp and ptc. 3. A-type cells beyond the range of Hh express ptc at a basal level [2115], and Ptc (by a double-negative route) relieves repression of PKA. PKA then phosphorylates Ci to provoke proteolysis of Ci’s transcription-

107

activating (dCBP-binding) domain, and the remnant (Ci-75) keeps hh, dpp, and other target genes off. The en-off state of A cells is inherited from the embryo [1093, 1124, 1790, 3129] (cf. Fig. 4.4). It is maintained by Polycomb [582, 2889, 4327], polyhomeotic [2728, 3512, 3753], dRaf [2532], fat-head [4607], and groucho (D/V boundary only) [998], so no further control by Ci-75 is needed. Hence, the Ci-vs.-En circuit is asymmetric: En keeps ci off in P cells, but Ci-75 does not keep en off in A cells (ci null clones in the A region do not turn en on [1078, 2832]). The en-on state of P cells is maintained by trithorax [452], moira [475], polyhomeotic [2728], and the EGFR pathway [207], rather than by the autoregulatory “En en” loop seen in embryos [1790]. Indeed, En en in discs [207, 1647, 4229], although this GOF effect may be spurious relative to the circuit’s normal range of operation [156]. It is possible to artiﬁcially drive ci expression (via act5c-Gal4:UAS-ci) in en-on (P-type) cells so as to make them co-express both Ci and En [943]. Such schizophrenic cells “hear themselves talk” (via “Ci-155 Ptc” and via “En Hh”) and behave like border cells [943]. In contrast, deaf-mute (en null ci null ) cells behave uniquely (vs. A or P or border cells) in terms of adhesive afﬁnity (cf. Ch. 6) [943]. Other questions pertain to Ci-75 vs. Ci-155, which have identical DNA-binding domains [2980]: 1. Because hh is a target gene of Ci-75 [2832], Ci-155 should turn hh on at the border [155], but it does not. Evidently, hh is more sensitive to inhibition by Ci-75 than to activation by Ci-155 (both of which exist there) [2832], but how this selectivity is achieved physically is unclear [2980]. Oddities of this zone suggest that other factors might override these controls [2728, 4478, 4539]. For instance, ectopic Ci-75 can repress ptc there but not dpp [155]. 2. A minimal amount of ptc transcription in A cells is needed to keep dpp off beyond the border zone [642], but if this baseline were due to a constitutive promoter, then ptc should also be expressed in the P region. So why isn’t it? Apparently, ptc’s expression in A cells stems not from its promoter per se but from an unknown A-type activator aside from Ci-155. Because anterior ptc expression does not increase when ci is removed [2832], ptc must be selectively “deaf” to Ci-75 (just as hh is deaf to Ci-155). Again, the basis for such selectivity is obscure [2980].

heteromeric receptor

Tkv

ks?

P

? enz microtubule

?

P

? P

Fu

cog

?

PKA

Cos2

primes for cleavage

~kinesin P

tr. factor Ci

cyt.

Slimb

P

cyt. . nuc

APC

dCBP co-activator

Arr

Dsh enz

enz

Sgg

PP2A P

cog cog

repressor

dCBP Brk

activates

P

dAxin

cog

Shn

repressor activates Target genes

n

P

Medea

co-activator

uc.

?

enz

? PKC

Mad tr. factor

Dfz2 cog

P

P

co-activator?

?

? CK2

blocker

?

PP2A

Su(fu)

?

receptor

Dad

P

enz

?

enz

mas

?

ligand Wg

Slimb

junction

co-receptor

Punt

?

Smo

Ptc

receptor

?

c Wingless pathway

primes for proteolysis

P

Shg

diverts into junctions

Arm co-activator

Target genes

Key binds activates inhibits phosphorylates de-phosphorylates

P

P

Dally

ligand Dpp

ligand Hh

Dally

b Dpp pathway

co-receptor?

a Hedgehog pathway

co-receptor?

108

Pan tr. factor

Gro co-repressor

activates

Target genes

CHAPTER FIVE. THE LEG DISC

Hh elicits expression of Dpp and Wg along the A/P boundary In 1994, the Short-range Inducer Model was invoked to explain Hh’s effects in leg discs. Hh was shown to turn on wg in the ventral half of the border, and to turn on dpp in both the dorsal and ventral halves (Fig. 5.8) [231, 1037, 2466]. Ventral dpp transcription, which is relatively meager [2739], turned out to be a red herring from the standpoint of patterning since dpp null V clones, unlike D ones, are anatomically normal [1811]. For some unknown reason, the dpp [2739, 2954] and wg [177, 3317] transcription zones are shaped like sectors, rather than bands. The dpp-on domain is narrower than the wg-on wedge and hence is often called a “stripe” [487, 3747]. The width disparity could trivially be due to dpp turning on at a higher Hh threshold, but it might exist for other reasons (e.g., dpp-on cells adhering more tightly or dividing more slowly). The sectors coincide roughly with the leg’s prospective D (dpp) and V (wg) midlines. If, as argued below, Dpp and Wg are morphogens for specifying positions along the D-V axis, then this state of affairs would explain the Gaps Mystery. To wit, the reference line would not be the A/P compartment boundary per se as ﬁrst thought [904], but rather the slightly offset line that bisects the dpp and wg transcription sectors (Fig. 5.4c). Embryos also manifest Hh-dependent Wg expression (cf. Ch. 4) [3386], but Dpp is not under Hh control at that time. In comparing a Wg blastoderm stripe to

109

the Wg sector of the leg disc (and doing the same for En [3747]), the leg disc seems equivalent to a body segment (or part thereof) that has curled itself into a circle [880, 1039]. This basic topology agrees with both the PC and Boundary Models, as well as with the Angular Values Hypothesis [2291]. Viewing the nascent disc as a circularized segment is appealing because it provides a simple way to co-opt the established intercellular signaling machinery of the segment polarity genes for circumferential patterning in the discs. [834]

The leg disc also resembles a body segment insofar as Dpp apparently acts as a dorsalizing agent in both systems (see below) [1807]: in the embryo, a Dpp gradient speciﬁes diverse fates within the dorsal 40% of the ectoderm [127, 1091, 1211, 4613]. Why should Hh turn on different genes in different parts of the A compartment? The answer is that dorsal (dAC) and ventral (vAC) cells of the A compartment differ in competence (Fig. 5.4c). These biases are revealed when Hh is expressed ubiquitously: Hh then activates dpp throughout dAC and wg throughout vAC [231, 617, 1037, 2466], rather than just along the A/P boundary (Fig. 5.9c). Activation of dpp in dAC (and weakly in vAC) and wg in vAC is also seen in somatic clones where the Hh pathway is turned on downstream of Hh by LOF mutations in patched [2058, 2491] or DC0 [2058, 2059, 2491, 2533, 3238]. For slimb LOF clones, the situation is more complex because slimb also functions in the Wg pathway [2060, 2856, 4279].

FIGURE 5.6. Some key signaling pathways in disc development. In the working models depicted here, certain components may

be replaced by others (not shown) in different tissues, regions, or time periods. Three cells are depicted, with apical microvilli at top, although receptor-ligand interactions may actually take place elsewhere on the surface. Black rectangles = proteins; gray area = cytoplasm (“cyt.”); dashed line = nuclear membrane. Arrows (activation) and blunt “ ” lines (repression) indicate epistatic relations. Noncovalent binding is symbolized by interlocking hooks (cf. key). Question marks concern how or whether the interactions occur (see text). A “cog” (a.k.a. “adaptor”) is a component that plays a steric (nonenzymatic) role [441, 1954, 3299, 3676] (e.g., Slimb [2668]). Enz = enzyme, CK2 = Casein Kinase 2, PKA = Protein Kinase A (likewise PKC), Shg = Shotgun (a.k.a. dE-cadherin). For other abbreviations, see App. 6. For perspective, see App. 7. a. Hedgehog pathway [1554, 1974, 2075, 3685, 3831]. Binding partners (listed from surface to DNA) are based on ﬂy studies or vertebrate homologs (vh): Hh-Ptc [755] , Ptc-Smo (vh [4130]), Fu-Cos2 [4068] , Fu-Ci [4068], Fu-Su(fu) [4068] , Cos2-microtubules [3612, 3976], Cos2-Ci [4068, 4536] , Su(fu)-Ci [4068] , dCBP-Ci [54], Ci-DNA [751] . Ci’s interaction with dCBP may be inhibited by PKA (link not shown) [54]. b. Dpp pathway [959, 1983, 3498]. Dally is placed upstream of Dpp, based on the ability of extra doses of dpp+ to suppress dally LOF phenotypes [2004], and a similar argument applies to Wg [2556] . Evidence for binding: Dpp-Punt [2495], Dpp-Tkv [2495] , Dpp-proteoglycans (e.g., Dally) [2004] , Punt-Tkv [2495], Tkv-Mad [1983], Tkv-Dad [1983], Tkv-PP2A (vh [1627]), Mad-Medea [1983, 4703], Mad-dCBP [4534], Mad-DNA [2217]. Mad may also bind microtubules (not shown) [1084]. The “Shn brk” link does not actually occur at the protein level, but rather occurs via inhibition of brk transcription [2727]. c. Wingless pathway [422, 3919]. Arr is thought to be a co-receptor for Dfz2, but it does not “present” Wg in the same way as Dally [4570]. Conceivably (not shown), Arr may act via CK2: Dfz2 Arr CK2 Dsh, etc. The APC homolog here (labeled simply APC) is likely E-APC [4832]. Sgg is thought to be Arm’s natural kinase [3683] (but see [3228, 4062]). Evidence for binding (see [3316] for details): Wg-Dfz2 [310], Wg-proteoglycans like Dally [598] , CK2-Dsh [3149, 4676], Dsh-dAxin (vh [2238] ), Sgg-APC (vh [3675]), Sgg-dAxin [3683], APC-dAxin [1698], PP2A-dAxin (vh [1919]), dAxin-Slimb (vh [2241]), APC-Arm [4832], dAxin-Arm [1698], APC-Arm [4832], Slimb-Arm (vh [4701] ), Shg-Arm [3749, 4268, 4416] , Arm-Pan [519, 4439], Gro-Pan [692, 2496], Pan-DNA [519, 4439].

110

IMAGINAL DISCS

Leg disc

a

Wing disc

D A

P V

dorsal region

b

A P En

c border zone

Smo

Ptc

Smo

PKA

ptc

PKA

Hh

hh

diffuses

ptc

(low)

P

Ptc

en

En

P P P P P

Ci-155 e ot pr

Ci-75

om as

Zn

dpp

Zn

dCBP

e

dpp dpp

d en ci dpp

hh omb

diffuses

Dpp

diffuses

omb

ci

CHAPTER FIVE. THE LEG DISC

The dAC/vAC border is not a strict lineage boundary [487, 1800, 4076], so dAC-type cells can presumably jostle into vAC territory and vice versa. No genes have been identiﬁed that are expressed strictly within dAC or vAC [3919], nor are any genes transcribed along the dAC/vAC border (as expected based on Boundary Model II) [4277], so the nature of these areas is not obvious [617].

Dpp dorsalizes and Wg ventralizes, or do they? Conceivably, dAC and vAC states are maintained by the Dpp and Wg signals themselves [487, 2954, 4277]. That is, AC cells within range of the Dpp signal adopt dAC competence, while those within range of Wg adopt vAC competence. Such a system could be stable without celllineage restrictions (cf. the Cabaret Metaphor). Support for this idea comes from an experiment where cells are tricked into switching states: when dAC cells are duped

111

(via sgg LOF ) into thinking they are receiving a maximal Wg signal, they stop making Dpp and activate their own wg gene whenever they are within range of Hh [1833, 2059] (notwithstanding a report to the contrary [1039, 4487]). Additional evidence is provided by the fact that the dAC/vAC border (which runs along row 7 in the fate map; cf. Figs. 5.1c and 5.4c) coincides with the line that separates the Dpp and Wg “spheres of inﬂuence” (Fig. 5.5a): in dpp LOF mutants the tarsus lacks structures on the D side of “L2&7 ” (the line that runs along rows 2 and 7), whereas, in wg LOF mutants the whole leg loses structures on the V side of L2&7 (Fig. 5.5b-d) [1812, 3317]. L2&7 is also deﬁned by leg defects caused by dpp null or wg null clones [1811]. It divides the disc into two sectors that would measure ∼225◦ (D) and ∼135◦ (V) if bristle rows were evenly spaced around the leg (cf. a femoral septum at this angle [3867]), but in fact there is a larger space

FIGURE 5.7. Effects of Hh diffusion into the A compartment of thoracic discs. For gene abbreviations, see text.

a. Left leg and wing discs, with D-V axes of the central regions oriented vertically (cf. compass). A and P compartments are lightly or darkly shaded, respectively. The boxed areas of both discs obey the “Hh Dpp” link, but the ventral part of the leg disc obeys “Hh Wg” instead (not shown). b. Part of the disc epithelium that straddles the A/P compartment boundary (magniﬁed from a). Hexagons are cells (although packing is not really so orderly). Cells in the P region express the transcription factor En, while A cells do not. Zigzag line marks the anterior limit of the dpp-on zone. c. Circuitry of genes and proteins (abridged; cf. Fig. 5.6) that operates in each of the cell types delineated in b. The chain of events (right to left) starts with expression of en in all P cells. (P cells have Smo but not Ci or Ptc [1022, 2832].) En activates hh but represses ci and dpp [1647, 2832, 2980, 3747, 4227]. Hh diffuses ∼8–10 cell diameters into the A region [4136]. Binding of its receptor Patched (Ptc) triggers synthesis and release of Dpp, which turns on genes like omb (cf. Fig. 6.3) [2455, 3074]. The “Hh Dpp” link is mediated by the transcription factor Ci (horizontal bar). Full-length Ci (155 kD, 1396 a.a.) has 5 tandem zinc ﬁngers (Zn), a C-terminal domain that binds the co-activator dCBP, a proteasome cleavage site (zigzag line), 5 residues capable of phosphorylation by Protein Kinase A (vertical lines) and other domains (cf. text) [54, 2832]. A-type cells do not express dpp in the absence of Hh (left) because basal transcription of ptc [4136] supplies enough Ptc to derepress PKA (via Smo). Phosphorylation of Ci by PKA leads to Ci’s proteasome-dependent cleavage, and the N-terminal fragment Ci-75 acts as a repressor (due to an alanine-rich section [65]) of dpp and hh [2832]. Dpp and Hh are both diffusible (triangles denote gradients), but Dpp’s range is greater (cf. Fig. 6.3). Omitted here is the enigmatic transcriptional regulation of ci by the zinc-ﬁnger protein Combgap (Cg) [621, 4017, 4217]: Cg ci throughout the A region of wing and leg discs, but Cg ci in the P region of wing (but not leg) discs. Curiously, the En hh link is inoperative in the developing eye [4168]. d. Summary of on/off states (thick vs. thin lines) of en, ci, and dpp. The dpp-on stripe is evidently “painted” in two ways: (1) directly via “Hh dpp” and (2) indirectly by a ﬁgure-ground “illusion” [657, 1585, 3747], where “Ci-75 dpp” anteriorly and “En dpp” posteriorly. Both actions stem from the disc’s en-on/off duality. Validation of this logic comes from larvae carrying an en transgene driven by a heat-shock promoter. When such larvae are heat-shocked, every cell transiently becomes en-on (P-type) and the dpp-on stripe disappears [3747]. A similar shut-off is expected for discs devoid of En, but this condition has not yet been created [1839]. Not shown is a trespass of the en-on edge into the border zone that occurs in late 3rd instar when a “Hh en” link pervades the disc [1647]. (Its ubiquity is shown by the ability of ptc null A clones turn on en [4136].) N.B.: Events in the boundary zone are more complex than depicted here (cf. Figs. 6.3–6.5), especially for Ci [4540]. En’s inhibition of dpp (En dpp) occurs both directly (by binding of En to dpp’s cis-enhancers [2980, 3747]) and indirectly via En’s suppression of ci (because Ci-155 dpp [1827, 2832]). Suppression of ci is also partly mediated by Polyhomeotic [3512] – a member of the Polycomb Group of repressors (cf. Ch. 8) [1012, 1867, 2374]. En’s activation of hh (En hh) is probably indirect (En X? hh) [224] because En normally acts a repressor [2047, 2048, 3175, 3860, 4003] by recruiting Groucho [1243, 2064, 4351] (but see [1706, 2642, 2729, 3481]). “X” is probably not Ci-75 because ci null clones in the A region do not turn on hh at P-type levels [1078, 2832, 2834]. Alternatively, En might turn on hh by using a co-activator, such as Blistered [2729], Extradenticle [3324, 3325, 3861], or the Trx-G chromatin-remodeling complex (cf. Ch. 8) [451]. En is potentially secretable [811, 2085, 2661] but is not used thusly here. Slimb helps keep wg off in the P region (not shown) via a Hh-independent route [2856]. After [2832, 4479, 4539]. See [2834] for further nuances.

112

IMAGINAL DISCS

5 6

4

D A

7

Hh 3

en 3-A

ci 3

dpp 3-P

Dpp 3

omb 3

brk 3

H15 3

hth 3

3

P

8

hh 3

V

2

h P

AN D

1

? ?

ac P

AND

wg E-P

Wg 3

? ? FIGURE 5.8. Sectorial domains of gene expression in the leg disc (which become stripes along the leg) plus one annular domain (hth). The ﬂow of control is from en (upper right), which establishes the A-P axis, through dpp and wg, which polarize the D-V axis, to ac (lower left), which allocates cells to bristle vs. nonbristle fates. Black areas chart mRNA (transcribed from endogenous or reporter gene) or protein levels during embryogenesis (E), late 3rd instar (3), early pupal period (P), or adult (A) stage. Protein levels are given for Hh (estimated based on data from wing discs [4136] ), Dpp (range is not exactly known [1169, 4265]), and Wg [2456] . Shading denotes degree of expression. Areas are mapped onto a disc, although some genes turn on only after 3rd instar (e.g., ac). The span of stages (e.g., “3-P”) indicates when expression has been seen but does not connote conﬁnement to that period. Wiring ( activation; inhibition) illustrates a few genetic interactions (see text or sources cited below). “and” means that both inputs are needed to activate the target gene. Thus, the dorsal hairy stripe is caused jointly by Dpp and Hh (neither alone is sufﬁcient), while the ventral stripe is caused jointly by Wg and Hh [1778]. Interactions between Dpp and Wg (not shown) involve mutual antagonism along the D-V axis and cooperation along the proximal-distal axis (see text). Genes (“DO” = details omitted): ac (achaete; DO: expression areas proximal to the tibia) [3193], brk (brinker) [2049], ci (cubitus interruptus) [1135], dpp (decapentaplegic) [2069, 3122] , en (engrailed) [2954] , h (hairy) [3193] , H15 (a T-box relative of omb) [3762] , hh (hedgehog) [4254] , hth (homothorax) [4760] , omb (optomotor-blind) [2049] , wg (wingless, whose stripe in pupal legs is ∼5 cells wide [1039]) [4277] . All these genes are known to have LOF or GOF effects on pattern, except H15 (whose null phenotype is wild-type [3087]) and omb. Two genes whose expression resembles that of H15 – nkd [4863] and Dfz3 [3977] (not shown) – also have no LOF defects. Enhancer-trap line c409 (not shown) matches ci’s pattern [2684], as does aldehyde oxidase [2340, 4051]. See [489] and FlyView (ﬂyview.uni-muenster.de) [2027] for other enhancer-trap expression patterns. Areas of expression that are not wedge shaped (omitted) include arcs (e.g., GAL30A [487], grain [509], Sex combs reduced [2393] , Ultrabithorax [3397]), parabolas [2977], a spot at the femoral chordotonal organ [1235], etc. [834, 1574, 2684, 4522]. Augmented from [531]. In some specimens the dpp-on and wg-on sectors appear as wide as shown [2954, 4277, 4849] , but often they look narrower [231, 487, 3762] . Likewise, the en domain sometimes displays a central acute angle as shown [2954, 3329] , but it can also seem straight [1472, 4254] or even obtuse [315]. These vagaries arise from variations in folding, viewing angle, and focal plane, as well as from inherent irregularities in the A/P (and other?) boundaries [1833, 4254]. In early pupae, a gap arises between the en domain and the dorsal h sector [1778], although it may not be as large as implied here. N.B.: Hh, Dpp, and Wg only seem to deﬁne D-V sectors of h expression [1778]. Therefore, one remaining mystery is: what factors deﬁne h-on sectors along the A-P axis? Another is: what factors deﬁne the edges of ac-on stripes that are not created by “h ac” inhibition?

CHAPTER FIVE. THE LEG DISC

between V rows 1 and 8 than between D rows 4 and 5 (Fig. 5.1e). Thus, the areas ﬂanking L2&7 may actually be closer in size – a detail that is important in assessing how far Dpp or Wg signals travel. Reduction of dpp or wg function does not just result in loss of structures from one or the other side of the leg. Commonly, the missing structures are replaced by a mirror-image copy of the remaining part (Fig. 5.5) [177, 1812, 3317, 4032]. The V/V dpp LOF phenotype is like a hand with two palms and no back, while the D/D wg LOF phenotype is like a hand with two backs and no palm. Both D/D [560, 1812] and V/V [2082] legs have uneven girths, suggesting that leg segments vary in the fraction of their circumference that is under wg or dpp control. According to the PC Model, such “Janus” oddities should, in principle, be easy to explain. The deﬁciency portion of the deﬁciency-duplication phenotype would be due to tissue loss, and the duplication would arise via juxtaposition of cells with disparate angular coordinates from around the necrotic zone, followed by circumferential intercalation by the shortest route. In fact, however, although necrosis is seen in dpp LOF discs [529, 2739, 2849], it is negligible in wg LOF discs [2015, 2929, 4683]. An even more serious problem for the model pertains to intercalation. According to PC logic, the V/V dpp LOF phenotype arises because loss of the D (dpp-dependent) sector removes ≥50% of the angular coordinates. If so, then the shortest route for row 2 and 7 cells to take when confronted must be the V arc that includes rows 1 and 8. However, the D/D wg LOF phenotype implies a different shortest route (i.e., the D arc encompassing rows 3–6) for the same row 2/7 confrontation. The PC Model is stumped by this paradox. When wg is artiﬁcially expressed in GOF clones on the D side of the leg, the cells in and around the clone make ventrolateral (row 2- or 7-type) bristles [837, 1039, 4159]. This result proved that Wg can act as a ventralizing agent [3923], but it raised the question of why the cells do not become fully ventralized. The failure to induce ventral (row 1- or 8-type) bristles was initially attributed to low output of the wg transgene that was used [4159], but two later ﬁndings revived doubts as to whether Wg is really a morphogen in the leg disc (cf. proof for such a role in the wing disc [4849]): 1. Activating the Wg pathway (cf. Fig. 5.6) downstream of the receptor (via sgg null clones) can induce 1/8type bristles in the very heart of the Dpp domain or anywhere else around the tarsus [1039, 1833, 4666]. This

113

fact refutes the excuse that Dpp is too strong in its own realm to permit V-type states there. Instead, it suggests that Wg only speciﬁes ventrolateral fates, with fates nearer the V midline being speciﬁed by a second ligand (another Wnt? [3919]) that is also transduced via Sgg [4666]. (Similar logic may govern the embryo’s Dpp gradient [2730].) 2. If Wg were a morphogen, then it should specify fates in a concentration-dependent way. However, when the disc is ﬂooded with Wg (via Gal4 drivers that enforce nearly ubiquitous expression of UAS-wg), the V region sprouts a forest of 2/7-type bristles but no extra 1/8-type bristles [4666]. Again, the implication is that Wg merely controls ventrolateral fates. Midventral identities might be dictated by Hh (cf. D rows [2533]), because extra 1/8-type bristles arise when the Hh pathway is turned on in the V area by DC0 LOF (Figs. 3d of [2491] and 6h of [3238]), although row 1 lies in the P compartment (deaf to Hh?) [1800, 2449]. Dpp’s role as a dorsalizing morphogen in the leg is equally dubious. Excess Dpp (UAS-dpp driven by dppGal4) can induce expression of omb (optomotor-blind ) – a marker gene for D identity – in the V region of the leg [487], but overexpression of Dpp within its own domain fails to induce extra (4/5-type) D pattern elements [2954]. The lack of a signiﬁcant GOF phenotype for a gene that has a strong LOF phenotype was seen before in scute (cf. Ch. 3). In that situation, the explanation was a combinatorial interaction between Scute and Daughterless – a dimerization partner that may be limiting under excessScute conditions. Here, too, with Dpp, there could be a simple reason why “shouting” achieves no more than “whispering.” Namely, another signal might be needed in addition to Dpp. Indeed, Hh might be the hidden player in the both the Dpp and Wg stories: {Hh and Dpp} {Hh and Wnt?}

row 4/5-type bristles? row 1/8-type bristles?

Dpp does display some dose-dependent properties (see below) [2954], but they are irrelevant to Dpp’s role as a dorsalizing agent per se. This confusing state of affairs stands in stark contrast to the embryo, where Dpp’s morphogen properties are so well documented in terms of fate speciﬁcation [1281, 3532, 3693, 4058, 4492], gradient distribution [1212, 1927, 2003], and dose sensitivity that is manifest not only in discrete thresholds [127, 1091, 1211, 4613], but also in haplo-insufﬁciency [1872, 1987, 3458, 3693]. It also contrasts with the wing, where Dpp’s role as a morphogen has been well established (cf. Ch. 6) [2455, 3074].

114

Dpp and Wg are mutually antagonistic In 1996, a ﬂurry of articles revealed a mutual antagonism between Dpp and Wg signaling. By manipulating dpp or wg expression directly [487, 2059, 2082, 2954, 4277] or by interfering indirectly with components of their pathways [2059, 3329, 4277], these experiments uncovered a ﬂip-ﬂop circuit that was suspected from earlier clues [1812, 2058]. Evidence that Wg inhibits Dpp is summarized below. 1. Under wgGOF conditions (dpp-Gal4:UAS-wg), where wg is forced to be expressed in the Dpp sector (i.e., dpp-expression zone), the level of dpp transcription decreases [487, 2082, 4277]. 2. In wg LOF mutants, dpp expression increases in the Wg sector (as wg expression decreases) [487]; this substitution is noticeable within 24 h after reducing wg function [4277]. 3. When wg expression is extinguished in the Wg sector (due to dpp-Gal4 activating UAS-dpp along the entire A/P border), expression of dpp-lacZ increases in that V area [487, 2954]. The same effect is seen in Wgsector somatic clones whose Wg pathway is blocked (by dshnull [1833, 2059]). 4. Somatic clones that are both DC0null (which makes cells think they have received a Hh signal) and wg null express dpp when they arise anywhere within the vAC (or dAC) [2059]. Evidence for a reciprocal inhibitory effect of Dpp on Wg is as follows: 1. Overexpressing dpp along the A/P boundary (dppGal4:UAS-dpp) suppresses Wg in the Wg sector [487] except at the disc center (endknob) [2954, 4277]. Concomitantly, expression of the enhancer trap H15-lacZ (a V marker) vanishes in the V region, and omb expression (a D marker) is now detected [487] (see Fig. 5.8 for expression domains). A similar effect is achieved when a constitutively active (ligand-independent) Dpp receptor (tkvGOF ) is used instead of excess Dpp (dpp-Gal4:UAS-tkvGOF ) [4277]. 2. When the level of dpp transcription decreases in the Dpp sector (due to dpp-Gal4:UAS-wg), the endogenous wg (distinguishable from the UAS-wg transgene by a lacZ insert at the wg locus) is activated [487]. (Concomitantly, H15-lacZ expression appears and omb expression disappears [487].) Upregulation of wg also occurs (although only rarely) when dppnull clones reside in the Dpp sector [2059] or when Dpp transduction is blocked in dorsal somatic clones (by puntLOF [3329] or tkvLOF [2059, 3329]).

IMAGINAL DISCS

3. In dppLOF mutants, Wg is now detected in the Dpp sector (as dpp expression decreases) [487]. The ectopic Wg is restricted to the tarsal region – the only area converted to V identity in these hypomorphic genotypes [1812]. A similar substitution of wg for dpp expression occurs when the Dpp pathway is blocked ubiquitously (by puntLOF [4277]), but in that case wg is transcribed in the entire dorsal sector [4277]. 4. Somatic clones that are both DC0null and dppnull express wg when they arise anywhere within the dAC (or vAC) [487, 2059]. Similar effects are seen when DC0null is coupled with sggnull , which inhibits dpp